• Anthropic's Boris Cherny says companies are right to focus on their ROI for AI.
• At the same time, Cherny said employees at all levels and roles still need tokens to be able to experiment with AI.
• Then, the Claude Code creator said, companies can start to control costs.

Claude Code creator Boris Cherny has a message for companies that are nervous about their AI token budgets.

"ROI is absolutely the right framing because you don't want to just think about cost because you kind of spend something on it and you get something back," Cherny said during a recent fireside chat at Scale AI.

Jesse Chen, Meta's director of product management who moderated the chat, asked the Anthropic employee directly about the recent concerns raised by Uber COO Andrew Macdonald about whether the rideshare giant's AI spending was leading to enough of a return to justify the rising cost of AI tokens.

Tokens are units of text that serve as a measurement for AI usage, such as the prompts processed by large language models, including those that power chatbots like Anthropic's Claude or its generative AI coding tool, Claude Code.

Cherny said it's right to be focused on ROI. It's also important, he said, not to overdo it in response to cost concerns.

"The way to do this is give people tokens and give them safety to experiment so they feel like they can try stuff and they're not going to get penalized for it," he said. "Once you find these internal use cases that kind of work, then you want to control the costs and you want to do that on the backend, not on the front end."

Otherwise, companies might miss out on the best ideas for deploying AI.

"Often, some of the most interesting ideas and the most innovative ways to improve processes and new product ideas are going to come from an accountant somewhere in the corner of the org or a marketing person that the CEO has never heard of," Cherny said.

Cherny emphasized that Anthropic offers several ways for its enterprise customers to control costs and set budgets, including per-seat cost controls.

Others in the AI space, including OpenAI CEO Sam Altman, are also increasingly discussing companies' concerns about the ROI of their AI investments.

As Cherny mentioned, AI firms like Anthropic are essentially token generators. That also means that they have an incentive to keep selling their models and generative AI tools, especially as they approach highly anticipated IPOs. The creator of Claude Code said that Anthropic is also paying attention to how its tokens are used.

"They're not free for us because every token we use is a token we do not give to a customer, so there's an opportunity cost," he said. "When I think about it, it actually maybe comes back to ROI."

Measuring that ROI is also changing, Cherny said, as the pace of AI model advancements continues to accelerate. He previously said that companies may have looked at the percentage of code written by AI. Cherny said that measurement is no longer as useful once more people let AI write 100% of their code, as he does.

"Then think about, how much is the code per engineer accelerating? And then the third thing to think about is like, what are the other bottlenecks that are getting in the way?" he said. "Because once you get it to this point where engineers are just writing a lot of code, the bottleneck is going to be like good ideas. So, how do you un-hobble that so that your company can generate ideas faster?

• CFOs are taking charge of AI spending as companies pour billions into the technology.
• Some are introducing AI budgets and new controls to keep costs from spiraling.
• "The CFO is really becoming the face of the AI story," said a PwC advisor to finance chiefs.

At Match Group, every employee now has an AI budget.

The parent company of Tinder, Hinge, and other dating apps recently began giving department heads a set amount to spend on AI, which is then distributed across their teams. Employees can track their usage on a dashboard, and if they want to exceed their budget, they have to explain why. The company's most expensive AI models also aren't available by default and require a specific use case.

"If you don't set guardrails, there's no reason for an engineer to not go use the most expensive model," said Match Group CFO Steve Bailey. The average software engineer at the company spends roughly $600 a month on AI tokens, he said.

Match Group's system reflects a growing reality across corporate America: As companies spend billions on AI, CFOs are emerging as some of the most powerful executives in the AI era.

Finance chiefs are doing more than signing off on AI budgets. In many cases, they're the ones deciding who gets access to AI tools, how much employees can spend, which vendors make the cut, and whether AI investments are generating enough value to justify costs.

"The CFO is really becoming the face of the AI story," said Peter Pollini, a PwC advisor to finance chiefs in the financial-services sector.

A spending boom

The stakes are enormous. Match Group initially allocated $5 million for AI this year, but it's now on track to spend double that amount, Bailey said. The increase followed CEO Spencer Rascoff's May push to make the company more AI-native by expanding access to AI tools across the workforce. Initially, they were available mainly to engineers.

"Aside from maybe travel and entertainment, we've never had to budget for a cost that's this big at the employee level," Bailey said.

To help fund those investments, Bailey said Match Group plans to dramatically slow hiring while it assesses how AI could reshape its workforce.

Across corporate America, similar calculations are turning CFOs into the gatekeepers of one of the biggest spending booms in decades.

At Elevance Health, CFO Mark Kaye oversees a hidden way of keeping AI costs from spiraling. The insurance giant quietly routes employees' queries to different AI models based on the complexity of the request. That's because a single prompt can cost anywhere from a few pennies to more than a dollar, depending on how many tokens, or units of data, employees gobble up.

"We manage it on the back end," said Kaye, adding that he expects Elevance Health, the parent of Anthem Blue Cross Blue Shield, to invest $1 billion or more on AI this year.

Making AI pay off

Some CFOs are making tough decisions about how to fund AI expenses at large, such as by freezing annual salary raises and laying off workers. Others say AI is helping to pay for itself.

Kaye said AI automation at Elevance has reduced administrative work tied to medical-chart reviews by roughly 40%, giving staff more time to support customers.

"There are significant inefficiencies in the system that AI is allowing us to take out," he said.

Keeping a tight leash on AI spending isn't the only new hurdle for CFOs. They're also responsible for managing spending on a category that is evolving more rapidly than previous generations of enterprise software.

For the first time this year, Xero, a global small-business platform that offers accounting, payroll, and payments, added a line item to its budget for AI token spending per employee, said Claire Bramley, the CFO. The company also created a task force to review software purchases and identify AI products it can do without.

"Do we have more than one tool that serves the same purpose?" Bramley said. "As a CFO, you want to make sure that everybody's not going off and doing their own thing."

AI is also changing who CFOs spend time with. Bramley said finance, technology, and HR leaders at Xero now work together more frequently to evaluate software purchases, hiring plans, and how AI could affect future staffing needs.

"You could probably do it once a month before, and I think you have to do it weekly today," she said.

Additional headaches

CFOs are also facing new business problems arising from AI.

Netta Samroengraja, finance chief at healthcare platform Zocdoc, said her team has had to hustle to evaluate AI tool providers to solve problems that, ironically, were created by the technology. In recruiting, for instance, the technology suddenly enabled job seekers to flood the company with applications and create phony personas.

"It was pretty prevalent very quickly, and so we had to react quickly," Samroengraja said.

That wasn't the only surprise, as the economics of AI were shifting, too. Early on, Zocdoc raced to vet vendors, anticipating that prices designed to attract customers at the start of the AI boom would increase over time.

The company used that window to test multiple providers and compare their cost and effectiveness before settling on the tools that delivered the strongest business results, Samroengraja said, adding that Zocdoc has been willing to spend more on tools that produce measurable business outcomes rather than optimize for the lowest possible AI spend.

"If you see the ROI in it, you should keep investing in this," she said.

A crowded AI market is making those decisions even harder. New providers are constantly pitching tools that promise to boost productivity, cut costs, or replace existing software, forcing many CFOs to take a more active role in evaluating vendors, said Alex Sobol, cofounder of the Millennium Alliance, an invite-only community for C-suite executives in North America and Europe.

"It seems like every hour there's a new AI vendor," he said. "It's hard to know what's real and what's fake, and what's good and what's bad."

• Cooley is joining a sudden rush of law firms creating their own AI technology, with help from Legora.
• Cooley Go Lab is an online portal built to help startups with routine contract review and drafting.
• The portal will be available exclusively to founders in Y Combinator's summer cohort to start.

Law firms know more clients are asking chatbots for advice before ever calling a lawyer. So legal giant Cooley is building technology that it hopes founders will use instead.

Cooley plans to give select startups access to Cooley Go Lab, an online portal where founders can upload files and ask questions about their documents, Matt Bartus, global cochair of Cooley's emerging companies and venture capital practice, told Business Insider.

To build it, Cooley teamed up with Legora, a fast-growing legal technology startup that sells to law firms and corporate legal departments. Last year, Legora entered a new line of business with what it calls "portals" — white-labeled workspaces where law firms and their clients can work together on legal matters.

Cooley Go Lab is aimed at catching a common startup problem early, Bartus said. Founders often handle routine contracts themselves to avoid outside counsel's hourly rates. That can leave startups with a trail of messy agreements that their first in-house lawyer has to unwind later.

Legora founder Max Junestrand knows the problem well. When he started the company at age 23, he said he used an early version of ChatGPT to rewrite contracts. Junestrand, a software engineer, not a lawyer, said he let some early contracts include an unlimited liability clause — a provision that can leave a company exposed to damages far beyond the value of the deal.

"When our general counsel started, she freaked out," he said.

Cooley is now trying to give the next crop of founders a way to use artificial intelligence to move faster, but with a law firm's guardrails around it.

Cooley Go Lab will have a limited rollout to start. It will be available first to startups in Y Combinator's summer cohort.

Much has changed for startups since Legora's turn in the famed startup accelerator. Teams can write code and release technology faster with coding agents. They are signing customers and growing revenue earlier, and the hottest companies seem to be raising funding nonstop. But moving faster also means legal work that once came later in a startup's life is being pulled closer to the beginning.

At Y Combinator, partner Gustaf Alströmer is seeing that shift play out in real time. In the last batch, a record 14 startups reached $1 million in annual recurring revenue — the amount of revenue a company expects to collect over a year. Alströmer said giving founders access to tools like Cooley Go Lab could help them keep that pace without creating contract-slop.

The portal includes features that review documents like nondisclosure and contractor agreements and flag issues for founders to consider. The tool also draws on Cooley Go, the firm's central hub of standard startup forms, templates, and guidance.

If you can't beat 'em, join 'em

Law firms like Cooley are facing a new reality. The better the frontier models get at legal work, the more founders and in-house lawyers may decide they can handle situations themselves rather than send them to outside counsel. Anthropic is trying to make that easier with new tools built for contract review and drafting.

Law firms are now figuring out how to stay relevant as more legal work becomes software-enabled. Some are building their own tech. Kirkland & Ellis has tapped Palantir to help it build tools to manage parts of the firm's private funds practice, while Freshfields is working with Anthropic on software that could eventually be sold to other law firms. Harvey, a leader in legal technology, says it's getting into training custom models for law firms.

Taken together, the moves point to a new attitude spreading across Big Law: If clients are going to use artificial intelligence anyway, law firms want to shape how they use it.

Bartus doesn't seem especially concerned about losing business to in-house legal departments. Cooley has been on a hot streak. Profits rose 6.7% to $922 million last fiscal year. The firm also scored a fair use win for Meta in a copyright case involving its model Llama last year, and it advised Jony Ive's hardware startup Io in a $6.5 billion sale to OpenAI.

Bartus is confident that companies will continue to depend on law firms for the important work. Cooley Go Lab, he said, is meant to help founders handle routine legal work more cleanly, not replace lawyers.

Cooley says the tool is not protected by attorney-client privilege, so founders will need to be careful about what they upload because those materials could be turned over in litigation.

"If you want actual legal advice," he said, "you need to talk to a lawyer."

• The age of freewheeling AI spending may be coming to an end.
• Consulting firms are rethinking how much they, and their clients, spend on AI.
• Tell us how spending at your consulting firm has changed.

Companies are learning that there's such a thing as spending too much on AI.

As the cost of AI tools grows, executives are recalibrating. Amazon recently removed its employee-made leaderboard for tracking AI token usage because it encouraged excessive spending. Walmart, which developed a vibe-coding tool for employees, recently set limits on the use of tokens. Uber COO Andrew MacDonald said it's hard to justify the money his company is spending on AI.

Cisco Chief Product Officer Jeetu Patel also pushed back on the cost of tokens. He said at an event recently that the price is "far higher than the actual value these tokens are generating at scale."

For the consulting industry, the rise of AI was a near-existential threat. At first glance, chatbots can do a lot of the work of consultants, particularly those early in their careers. Most firms moved quickly to attract clients who needed help integrating the technology into their own companies. And they quickly adopted it themselves.

KPMG, for example, has built a dashboard to track how often employees in its US advisory division use AI tools, part of a broader effort to move from basic adoption to more sophisticated use. McKinsey plans to go further. CEO Bob Sternfels said in January that the firm uses roughly 25,000 AI agents alongside its 40,000 human employees, and hopes one or more agents will eventually support every employee.

The surge in spending, however, has raised a question: Are companies investing in AI strategically or simply spending to avoid being left behind? It's something consulting firms are working to answer for both their clients and themselves.

Tell us how AI spending has changed at your consulting firm:

For now, the answer appears to be: keep spending, but more strategically.

In a recent report on corporate AI investment, Boston Consulting Group found that companies expect to more than double their AI spending in 2026, from roughly 0.8% of revenue to about 1.7%. For large enterprises, that shift represents billions of dollars flowing into AI strategies that remain, in many cases, experimental and difficult to measure.

Russell Fradin, CEO and cofounder of Larridin, a platform that helps companies — including major consulting firms — measure the returns on AI usage, said the spending trend will continue.

"We haven't seen anyone talking about spending less in AI next year," Fradin told Business Insider. "They're just talking about instrumenting to understand where it goes."

Companies, Fradin said, are coming to the consensus that they "can't 10x spend every year forever."

Em uma trajetória pouco comum em suas negociações no mercado de capitais deste ano, as ações da Alphabet registravam queda de 5,46% na Nasdaq por volta das 14h20 (horário local), cotadas a US$ 347,39, avaliando a holding em US$ 4,24 trilhões. O ponto de partida para esse recuo dos papéis é a disputa cada vez […]

O post As derrotas da Alphabet na batalha por talentos de IA apareceu primeiro em NeoFeed.

A Microsoft foi uma das grandes propulsoras de inteligência artificial generativa no mundo. Agora, Satya Nadella, o CEO da companhia, está assustado com os rumos que a tecnologia está tomando. Em entrevista ao The Wall Street Journal (WSJ), o executivo fez duras críticas à forma como a corrida pelo domínio da IA tem se desenvolvido, […]

O post CEO da Microsoft muda o discurso e critica gigantes de IA após perder terreno para Anthropic e OpenAI apareceu primeiro em NeoFeed.

Brasileiros que vivem fora do país ganharam uma nova opção para manter a documentação em dia. Desde esta quinta-feira (18), o Consulado-Geral do Brasil em Assunção, no Paraguai, passou a emitir a Carteira de Identidade Nacional (CIN), dentro do projeto do governo federal iniciado em abril, em Lisboa (Portugal) e expandido para a América do Sul.

O Paraguai foi escolhido por concentrar a terceira maior comunidade brasileira no exterior, com mais de 250 mil pessoas, atrás apenas dos Estados Unidos e de Portugal. O crescimento dessa população está ligado a fatores como trabalho, estudos e investimentos no setor agrícola.

O solicitante deve enviar certidões de nascimento ou casamento para validação prévia. Após análise, o agendamento presencial é liberado no órgão diplomático brasileiro na capital paraguaia.

No consulado, a coleta de dados biométricos, como impressões digitais e fotografia, é feita em totem eletrônico, com apoio remoto. Depois da validação, o documento é produzido no Brasil e enviado por malote diplomático para a entrega ao cidadão.

Em Lisboa, onde o projeto começou, mais de 400 carteiras de identidade foram emitidas em dois meses, segundo o Ministério da Gestão e da Inovação em Serviços Públicos (MGI). No Paraguai, a expectativa é ampliar esse número nos próximos meses.

Sem passaporte em quase toda América do Sul

A CIN substitui o antigo RG e utiliza o CPF como número único de identificação. O modelo busca reduzir fraudes e unificar os registros civis no País. O documento também facilita a vida de brasileiros no exterior e pode ser usado em viagens a oito países da América do Sul sem necessidade de passaporte.

Além do Paraguai o brasileiro pode entrar na Argentina, Uruguai, Bolívia, Chile, Colômbia, Equador e Peru portanto apenas a CIN.

Outro impacto para quem possui a carteira é o acesso à conta nível “Ouro” na plataforma Gov.br, que libera mais de 5 mil serviços digitais, incluindo Carteira de Trabalho Digital, Meu INSS e Meu SUS Digital, sem necessidade de deslocamento ao Brasil para resolver demandas administrativas.

Quais documentos podem ser emitidos no exterior?

Além da nova carteira de identidade em fase de testes, a rede consular brasileira mantém serviços regulares para cidadãos no exterior. Entre os principais documentos disponíveis estão:

• Passaporte brasileiro, emitido em embaixadas e consulados
• CPF, com inscrição, regularização ou atualização cadastral
• Certidões de nascimento, casamento e óbito
• Autorização de retorno ao Brasil, para casos de urgência sem passaporte válido
• Procurações com validade jurídica no Brasil
• Atestado de residência no exterior, usado para fins fiscais e administrativos

Para solicitar documentos no exterior, o cidadão precisa fazer agendamento prévio pelo sistema e-Consular, plataforma do Ministério das Relações Exteriores (MRE). Não há atendimento direto sem cadastro.

*Sob supervisão de Gustavo Porto

• The US restricted Anthropic's new cybersecurity AI models, prompting the company to suspend all access.
• The restrictions are creating winners and losers in the red-hot AI race.
• Mistral and DeepSeek could benefit as sovereignty concerns boost the appeal of open-weight models.

One company's headache may be another's opportunity.

The White House's restrictions on access to Anthropic's new AI models have created winners and losers across the AI industry.

On Friday, US officials restricted access to Anthropic's cybersecurity-focused models, Mythos 5 and Fable 5, after concluding safeguards designed to prevent misuse of Fable 5 could be bypassed. The restrictions block foreign nationals from accessing the systems. In response, Anthropic shut down access for everyone.

The move has dealt a direct blow to Anthropic. But it may also strengthen the position of AI companies providing more open models that their customers can deploy and control themselves.

Here are the biggest winners and losers.

Mistral

The verdict: Winner.

Why: The French startup has spent more than a year making the case that Europe should not become dependent on American AI providers.

Unlike Anthropic, whose most advanced models are accessed through company-controlled systems, Mistral has championed open-weight models that customers can deploy on their own infrastructure and customize using their own data.

The Anthropic restrictions gave its CEO, Arthur Mensch, a real-world example of the risk he has been warning about.

In an X post on Tuesday, Mensch doubled down on Mistral's sovereignty pitch, saying the company's upcoming models would be open-weight because users should be able to "own, inspect, audit, or improve" the AI systems they use.

The timing couldn't be much better for Mistral.

France announced this week that its domestic intelligence agency would replace Palantir's AI data tools with those of a French provider, with Prime Minister Sébastien Lecornu warning against "strategic dependencies" on foreign technology.

DeepSeek

The verdict: Winner.

Why: Like Mistral, DeepSeek's open-weight approach may suddenly look more attractive.

Unlike Anthropic's Mythos 5 and Fable 5 models, which are controlled by the company, DeepSeek's models can be downloaded, modified, and deployed by customers themselves.

That makes DeepSeek a beneficiary if governments and businesses begin prioritizing control and sovereignty over access to the latest closed models.

The episode also gives China an opportunity to argue that reliance on US AI providers comes with geopolitical risks, giving it a boost in an increasingly narrowing AI race. During an Anthropic event last month, its CEO, Dario Amodei, said Chinese AI models were roughly 6 to 12 months behind leading US AI systems.

Anthropic

The verdict: Loser.

Why: Anthropic is the company directly affected by the restrictions.

The export controls block foreign nationals, including Anthropic's own employees, from accessing Mythos 5 and Fable 5, which limits the company's ability to distribute some of its most advanced systems internationally.

More importantly, the episode highlights a potential weakness of closed AI models. Because Anthropic controls access to its systems, governments can, in turn, exert greater influence over who can use them.

The episode is also the latest headache for Anthropic in a monthslong spat with the White House after the AI firm said its technology should not be used for domestic mass surveillance or fully autonomous weapons systems.

In response, the US government designated Anthropic a supply chain risk, then Anthropic challenged the move in court.

US AI companies with closed models

The verdict: Short-term winners, long-term losers.

Why: While the restrictions may give a short-term boost to Anthropic's US rivals, it also raises uncomfortable questions for them down the road.

Companies including OpenAI, Google, and xAI primarily distribute their most advanced models through platforms and services they control.

Meta is a partial exception. While some models in its Llama family are open-weight, the company has increasingly been moving towards closed models that it has tighter control over, such as Muse Spark.

For governments and businesses, the Anthropic episode serves as a reminder that access to AI can ultimately depend on decisions made by providers and the governments that oversee them.

That dynamic could strengthen the appeal of sovereign and open-weight alternatives in Europe and elsewhere.

• A neuroscientist worries some people are letting AI do too much of their thinking.
• Over time, she says, that could weaken cognitive reserve, a key defense against dementia.
• "How you use AI, not how often, will determine its impact," Vivienne Ming told Business Insider.

AI doesn't cause dementia, but how you use it could weaken one of the brain's core defenses against it.

That's the warning from Vivienne Ming, a theoretical neuroscientist, the chief scientist at the Possibility Institute, a metascience research group, and founder of Socos Labs, an AI and education firm.

"Your chatbot is not giving you Alzheimer's," Ming told Business Insider.

"My worry is the cumulative impact of chronic substitution: when you stop doing the cognitive work because something will do it for you, you stop building the reserve that protects you later," she said.

As AI has swiftly become an integral part of people's lives and careers, AI researchers and some tech leaders have been releasing warnings about its deskilling effect, the slow erosion of job skills, and the decline in independent thinking.

Ming went a step further, saying that repeatedly outsourcing mental effort to AI, especially among young people, could have real implications for long-term brain health.

"That's the group from whom I'm most concerned," she said. "How you use AI, not how often, will determine its impact."

Over the long term, Ming worries that routinely outsourcing thinking to AI could reduce cognitive engagement and make it harder to build cognitive reserve — the brain's ability to adapt and remain resilient in the face of damage or aging.

"The mechanism I'm describing is the classic 'use it or lose it,'" Ming said.

'GPT is the new GPS'

To drive her point home, Ming compared the effects of using GPS and an AI chatbot.

Researchers at McGill University in Montreal found in 2020 that people with greater lifetime GPS experience have worse spatial memory during self-guided navigation.

In a four-month small study conducted over four months last year, MIT's Media Lab found that people who used a large language model to help write essays showed weaker neural connectivity than participants who used search engines or no external tools, and often couldn't accurately quote passages from essays they had written minutes earlier.

These two examples, Ming said, are cases of cognitive offloading and surrender, or, as she put it, "delegating the effortful part of a task to an external system so your own networks never have to do it."

Her concern in both cases is that people may be engaging key brain functions less frequently, including the hippocampus, the part of your brain that is responsible for memory and learning, and the prefrontal brain networks that help with attention, self-control, and decision-making.

"The hippocampus and prefrontal networks doing that work are precisely the systems that matter for cognitive aging," she said.

"GPT is the new GPS," she added, referring to OpenAI's chatbot ChatGPT, which she said could erode cognitive skills if people increasingly rely on it to think for them.

A matter of cognitive reserve

Research has consistently linked mentally stimulating activities to higher levels of cognitive reserve and lower dementia risk.

One analysis conducted by the English Longitudinal Study of Ageing (ELSA) in 2020 on 12,280 adults aged 50 and older, found that older people with higher cognitive reserve can expect to have a 35% lower risk of developing dementia compared to those with lower levels.

"The principle that lifelong mental engagement delays cognitive decline is some of the most replicated research we have," Ming said.

Importantly, Ming said no biomarker study linking AI use to dementia pathology has been conducted yet. Most of the data right now is "correlational or short-term," she said.

However, she thinks now is the time to start analyzing this cohort, "while the behavior is still taking shape."

"By the time we have the dementia data, a generation will have already formed the habit," she added.

• Allbirds has been rebranded as Smartbird. Instead of selling shoes, it's an AI infrastructure provider.
• Nadia Carlsten, the new CEO, spoke exclusively to Business Insider about her plans.
• She wants to build custom, single-tenant AI infrastructure for mid-market enterprises.

A few years ago, Allbirds was Silicon Valley's favorite sneaker company. Now it's betting its future on AI infrastructure.

After selling off its footwear business and shedding most of its workforce, the company formerly known for its eco-friendly wool sneakers has reinvented itself as Smartbird, an AI infrastructure provider led by a CEO who has never worn its signature shoes.

The transformation is one of the most dramatic pivots of the AI boom and a test of whether a struggling public consumer company can transition into an AI company.

"I'm more of a high heels person myself," Nadia Carlsten told Business Insider in an exclusive interview. "I'm blissfully unaware of all things Allbirds."

For those unaware, Allbirds launched in 2015 and quickly became one of tech's hottest consumer brands, with its sneakers as much a part of the Silicon Valley uniform as hoodies and Patagonia vests.

After going public in 2021, Allbirds was worth nearly $4 billion. But the brand's cool factor faded almost as quickly as it arrived. By early 2025, Allbirds' market value had fallen below $20 million.

Then, in April, the company announced an only-in-2026 pivot: It would no longer sell shoes and instead become an AI infrastructure provider, going head-to-head with the likes of Amazon, CoreWeave, and Crusoe.

Some ridiculed the move as "bizarre," or even "ridiculous and concerning." Wall Street was more enthusiastic, with the stock briefly soaring 800% on the news, though it has since lost much of its gains.

The transformation became official on Wednesday as the company said that it has completed the sale of the Allbirds brand and footwear assets, changed its legal name to Smartbird, and appointed Carlsten as president and CEO. She replaces Joe Vernachio, who is resigning from the company and the board of directors.

Carlsten was previously CEO of the Danish Centre for AI Innovation. She also managed product portfolios at SandboxAQ and launched the quantum computing service at Amazon Web Services.

With nearly the entire company's staff gone, Carlsten is starting from scratch, except for the same BIRD ticker symbol that trades on the NASDAQ.

Business Insider spoke with Carlsten about her plans. This interview has been edited and condensed for clarity.

BI: Did you ever imagine you would be working at Allbirds, and it would not be a shoe company, but an AI infrastructure play?

Carlsten: I've been around the block in Silicon Valley. It's not that unusual. Slack started as a game. Twitter started as a podcast. SpaceX started as a rocket company and is now doing AI infrastructure. So there's precedent for some of this.

Everybody's trying to be in the AI infrastructure space. This is probably not the most typical way to get into it, but we have a really good plan and strategy.

In a few months, people won't even remember the shoes.

BI: What does the company you're taking over look like at this stage?

Carlsten: The important thing to remember is that the shoe business has been sold, so anybody who was dedicated to the retail business is no longer part of the company. My first task is to hire the team. This is a brand-new company with brand-new people.

BI: Why not just start a new company?

Carlsten: In many ways, it is like a startup. I'm going to be growing a team, developing a new business model, approaching customers, and growing a pipeline of customers.

There are also some advantages to being a public company. One of them is access to capital. We have an easier time as we're looking at acquisitions and partnering with others in the industry. The liquidity makes it a lot easier to recruit.

In AI, speed is key. So why would you want to do things more slowly if you can do it faster?

BI: You're also now running a public company from day one, which brings a lot of scrutiny. Are you worried about that?

Carlsten: AI fluctuates, whether it's public or private. This is a business that is never static. AI is moving incredibly fast.

Customers are demanding things very differently than they were just a couple of months ago. So I don't think that makes much of a difference in how we will build the business.

BI: What exactly is Smartbird going to do?

Carlsten: We are an AI infrastructure company, and what makes us different is that we are focusing on the mid-market, such as enterprises that are in the pharma space or financial services space, and also countries that are interested in sovereign AI or having regional AI infrastructure accessible to them.

All of these players are doing more AI. They have more needs for persistent AI infrastructure, but at the same time, for whatever reason, they cannot use or won't use the public clouds. They are very interested in making sure their proprietary data does not enter a shared multi-tenant infrastructure system.

Right now, their choices are either shared infrastructure or building their own. And most of the people that I talk to in this space want to do more AI, but that doesn't mean that they want to build that AI infrastructure. Right now, they're doing it because they have to.

BI: You're going up against Amazon, Google, CoreWeave, and a lot of other established players. How do you plan to compete?

Carlsten: We're not competing head-on with hyperscalers like AWS or Azure or even large neoclouds like CoreWeave. Those guys are very good at building massive-scale shared infrastructure, which is the opposite of what we want to be doing.

We will focus on customers who need AI infrastructure at a smaller scale. Usually, they want single-tenant infrastructure, something that looks like a GPU cluster that they own and can fully control without having the disadvantages of managing the stack themselves.

BI: Where are you getting the GPUs from?

Carlsten: We'll be sourcing from multiple vendors. One of the things we'll be offering customers is the flexibility to build something specifically for their requirements.

BI: Are you going to buy the infrastructure or lease it?

Carlsten: We will purchase the infrastructure and build it out on behalf of the customer. We're not building ahead of demand. We are building something for specific customers.

We can be a lot more agile than the bigger players.

• As Americans use AI more they're trusting AI companies less, recent Morning Consult surveys show.
• At the same time, the popularity of nostalgic products like Lunchables and Capri-Sun is rising.
• The trend suggests consumers want the comfort of familiar brands in uncertain times.

The faster things change, the more we gravitate toward things that stay the same.

It's no secret that AI is disrupting many aspects of modern life, but recent survey data shows that the more people use the tech, the less they trust it.

The latest numbers from market research firm Morning Consult, released Tuesday, found that AI was one of the least trusted categories among US consumers, with seven out of 10 major AI brands seeing year-over-year declines in their net trust scores.

Google's Gemini managed to buck the trend and improve its score by six points to lead the pack with a net trust score of 24.

At the same time, some of the brands that saw the greatest improvement in consumer trust scores were decidedly low-tech and high-nostalgia: Capri-Sun, Lunchables, Hot Wheels, and Mr. Pibb.

"What unites them is that they belong to a specific register of American memory: the brand landscape of childhood, before adult complexity set in," Morning Consult said in the report.

The company has tracked trust scores for nearly 600 brands since 2018 and found that Americans' overall trust in consumer brands is higher than ever in spite of, well, everything.

Other high-ranking brands — Dawn dish soap, Band-Aid wound care, Heinz ketchup — are "reliable if not particularly exciting" and have "eliminated surprise from the consumer relationship," the report said.

The report also highlighted Gap's return to popularity, which it attributed to a heavy adoption of Y2K aesthetics, harkening back to a comparatively less complicated (or at least slower-moving) era.

Americans' fondness for decades-old standbys stands in stark contrast to their feelings about AI companies, many of which are evolving with head-spinning speed and driving eye-watering financial valuations.

A more detailed Morning Consult report on AI published in May found that more than a third of survey respondents do not trust AI "at all" — roughly matching the share of people who have a positive view of the tech.

Among ten of the leading companies, Meta AI, Perplexity AI, and xAI saw the sharpest declines in overall trust ratings since last year. One in five respondents agreed with the statement that AI companies' products present "a real risk" of ending human civilization.

"In 2026, a significant portion of consumers are in anchoring mode: seeking brands they can count on when other parts of their environment feel unstable," Morning Consult said.

Pylon CEO Marty Kausas had to make a difficult choice: scale back token spending, or stomach a $1.4 million bill.

Kausas said that his AI software company was fast approaching 150 employees on its Anthropic plan earlier this month, a point where the bill would more than triple. That realization got Kausas to declare the era of unlimited spending over — and he decided to set ceilings for tokens, the units of data that determine how AI is priced, for some of his non-technical employees.

Pylon's VP of finance is now exploring "where we should set caps," Kausas said. "This is just the start."

Leaders like Kausas are weathering a massive workplace shift, as more workers learn to love improved AI tools. Over the past few months, usage has moved from something bosses felt they had to incentivize to something they had to limit, due to skyrocketing costs and the realization that unlimited spending didn't always yield meaningful results.

OpenAI CEO Sam Altman said earlier this month he was blown away by how fast the conversation around AI budgets had changed. At the beginning of the year, "people were totally happy with the amount they were spending," he said. Now, these costs are "a huge issue."

It's not just CEOs and CFOs navigating these new corporate dynamics. For rank-and-file software engineers, part of their job now involves advocating for the compute they need to succeed. Meanwhile, some managers have to barter for their team's tokens, pitching like "Shark Tank." And, to poach red-hot AI talent, hiring managers are guaranteeing candidates tokens to spend.

A cutthroat Hunger Games for AI compute is fast approaching, one where everyone — from the C-suite to junior developers — is a player.

Token-fever whiplash

Max Kan's official job title is "tokenomics analyst."

At the data provider SemiAnalysis, Kan helps build token models for hedge funds and hyperscalers. When I called Kan in May, he was bullish on the impact that deep token budgets could have on the workforce. "It's basically true for everyone that, if you have an employee that's making $100,000 a year, you can probably make them 2x more productive with $10,000 worth of tokens," he said.

Those were the days of tokenmaxxing, when companies sent their engineers diving into token pools like Scrooge McDuck. Companies encouraged token leaderboards, where those at the bottom of the rankings felt pressure to use more AI, and executives across a range of industries couldn't stop talking about it.

The word "tokens" was used in 129 earnings calls in Q2 of 2026, up from 57 calls the prior quarter, according to an analysis performed for Business Insider by business intelligence platform AlphaSense.

Within a matter of weeks, the belt-tightening began. Companies began putting AI budgets on a diet and setting token limits. Coinbase set a cap; so did Walmart. Amazon shut down its internal token leaderboard.

Kan still advocates for big per-engineer allowances — and wonders what workers will think of the rapid discourse shift. He worried that engineers would think: "My boss is adamantly pushing me to do one thing, then I did that thing, and now I'm getting yelled at because I did that thing too well."

"I would definitely feel confused and angry if I were an engineer in those positions," he said.

Leaders across industries — from financial giants like JPMorgan to media conglomerates like Disney — are working to develop cohesive, effective AI policies.

Some firms have always been anti-tokenmaxxing. The enterprise software company Pega is one of them. When I hopped on the phone in May with its CFO and COO, Ken Stillwell, he called the trend an "incredibly self-serving" narrative by the AI companies. His company didn't set numerical token caps, but it did throttle requests that would spend in excess.

When we spoke a month later, as the discourse shifted, Stillwell felt vindicated. "We're quite happy that we're one of many talking about this," he said.

AI spending also continues to soar

Technology and media companies spent an average of $66.29 per employee on AI in May, up from $58.84 in April, according to Ramp's AI Index.

Ara Kharazian, its lead economist, told Business Insider he expected this metric to keep rising, but he spotted early signs of tightening, such as increased use of model routers, which can help better manage costs.

Some companies aren't cutting AI budgets just yet, but they are thinking critically about head count. For instance, MindFort, a Y Combinator-backed AI startup, has six employees. Its CEO, Brandon Veiseh, said the company would've needed 20 employees pre-AI to reach its current scale. Where have those funds gone? Tokens.

"We have to weigh our token-to-people ratio," Veiseh said. "It's not something we think is particularly comfortable or a great feeling to say."

Even though token costs are expected to come down as AI companies like Google increasingly compete on price by offering smaller, more efficient models, these sorts of tradeoffs aren't likely to go away. Often, the cheaper a resource is, the more of it is consumed.

For now, companies are thinking more critically and sometimes taking strategic steps back — but they're hesitant to move too quickly. Kausas, Pylon's CEO, said he wants to prioritize making sure there's a return on investment — and avoid engineer backlash.

"If we told engineers that they were not allowed to use AI products, they would not work here," he said. "It would feel like you were in the Stone Age."

Dawn of the token Hunger Games?

As engineers increasingly learn they might have to battle for their token allocation, team infighting could grow.

Some have compared this to a survival-of-the-fittest scenario. "Coding is now cockroach protein bars and we're all fighting for crumbs," said one coder on X, comparing the dynamic to "The Hunger Games."

Developers are also asking more about tokens during job interviews. Kausas said that applicants had asked him about budgets. AI advisor and AWS alum Allie K. Miller had heard of interviewees getting into the nitty-gritty: "What tier of model will I have access to? Do you have partnerships with AI labs that get us relatively early access?"

It's a sign of a new era where tokens — or at least the number workers want — aren't guaranteed.

Max Christoff, the CTO of legal tech company Everlaw, made the case for giving engineers token caps, but letting them negotiate for bigger budgets. He compared it to using cellular data before unlimited plans. Sometimes you need to spend big on the data, but other times you mindlessly scroll, not realizing how much you're wasting. Christoff wanted all of the former and none of the latter.

"We want to make it easy to ask for more if you can actually use it," Christoff said.

If a company doesn't set token caps, it may also set model restrictions. Russ Fradin, the founder of Larridin, a platform for tracking AI use, was emphatic. "Of course, they will limit who gets to use these tools. It's not even a question," he said.

Fradin compared allocating model access to taking a trip on the company dime. Many are allowed to book an economy flight, but few — if any — are allowed to charter a jet, he said. Access to cutting-edge AI models may be equivalent to the private jet: so expensive that only a few all-stars can do it.

Engineers have good reason to fight for their tokens. Having limited AI access could hurt them in the long run, leaving them less skilled or less marketable in future job searches.

Brock Simon advised companies on AI for Bain & Company before he founded his own startup, Native. He watched as some companies were slow to adopt the technology or restricted access to specific tools and agents, leaving their employees behind the curve.

"It really hurt some people's careers," he said.

• Billions of dollars are riding on the promise that artificial intelligence can absorb legal work.
• Crosby, a tech-driven law firm, built a benchmark to measure how well models negotiate contracts.
• Redline Bench is meant to help lawyers answer whether they can trust the technology's work.

Legal technology wants its vibe-coding moment. But first, it has to prove the tools can think like a lawyer.

Taking up the task is Crosby, a startup-meets-law-firm that sells basic legal services to companies, including Cursor and Rogo. On Wednesday, it released the Redline Bench, a tool built to measure how well artificial intelligence models perform real-world legal tasks, starting with contract review.

Software engineers have spent the past few years watching these systems get shockingly good at writing code and debugging errors. Now legal tech companies are chasing a similar prize: artificial intelligence that can review contracts, spot risks, and haggle terms faster and cheaper than lawyers.

But law has a problem that coding does not, says Ryan Daniels, a former in-house lawyer turned Crosby founder. "It's really hard to define 'good' or 'bad,'" he said.

Models can write code that either runs or breaks. Legal work is a murkier target. A sales contract can be edited, or "redlined," in lots of defensible ways, Daniels explains. A change that one lawyer sees as prudent, another might call too aggressive.

That ambiguity has become a headache for companies racing to automate legal work, from the scrappy neofirms to the model labs themselves. Anthropic has spent the past few months courting in-house lawyers with tools built for them. That push has been closely watched by investors. Earlier this year, Anthropic's new legal plugin stirred a sell-off in legal tech stocks.

Benchmarks are one of the main ways companies track progress. The labs building frontier models use them as stress tests, measuring whether a new system is better at tasks than the last one.

Coding has hundreds of benchmarks for evaluating models. But the legal industry still lacks a shared way to answer the question: Is the AI's work any good?

Crosby has been working on a new yardstick. The company pulled its engineers and lawyers into a tactical unit called Crosby Intelligence to build agents for Crosby's law firm and a benchmark to grade them against. That team includes engineer Sharan Ramjee, who worked on transformer models to sniff out fraud at Stripe, and Ross Weiser, a lawyer who joined from elite law firm Sullivan & Cromwell.

Crosby also partnered with Micro1, a company that helps model-makers recruit expert workers, to find more lawyers who could help define what counts as good legal work.

To build the benchmark, senior lawyers simulated software deals and marked the contract changes they considered most important at each stage of the negotiation. Those changes were turned into weighted criteria.

When Crosby runs a new test, it gives models the same contracts and asks them to make their own edits. Then a panel of three judges compares these redlines with the lawyer-built rubric. The judges vote pass or fail on each item, and the final score shows how often the models made the kinds of edits that lawyers considered important.

Redline Bench will be made public so any lab can put its models through Crosby's paces. Crosby also plans to regularly release reports tracking how major models compare.

The first release of the Redline Bench put ChatGPT 5.5 at the top of the heap, with a score of 50.5%, meaning the model's redlines matched half of the edits that lawyers prioritized. Gemini 3.5 Flash followed at 45.1%, and Claude Opus 4.8 scored 44.4%.

Crosby was able to test Anthropic's highly capable new model, Fable 5, only once before Anthropic pulled it off the shelves. The results were promising, with a score of 47.3%. When access is restored, Crosby will run the benchmark again and update it.

Crosby isn't the only company trying to measure how the models stack up. Harvey, one of the best-funded legal startups, has released benchmarks for case law research and contract review.

Anthropic and OpenAI also build their own benchmarks to measure performance on real-world tasks. But Daniels said those results can be hard to trust. Over time, the labs eventually tune their systems to perform well on their own tests, he said.

The stakes are bigger than a scoreboard. Billions of investment dollars are riding on the promise that artificial intelligence can lower legal bills and absorb work that used to pile up on the general counsel's desk.

Lawyers will only use the tools if they trust them. Crosby wants to give them a reason to.

• UK-based business owner Sean McDonnell relies on AI for his web design business and SaaS website.
• The White House ordered Anthropic to cut foreign access to Fable 5 while McDonell was mid-task.
• McDonnell emphasizes importance of backup plans due to AI tool disruptions like the Fable incident.

This as-told-to essay is based on a conversation with Sean McDonnell, 43, who lives in England. McDonnell is the founder of the web design company Kaizen and the SaaS website Consigns. The following has been edited for length and clarity.

Developing my website would not have been possible without AI.

I started my web design business earlier this year, which also led me to create a website that provides software to help companies track their waste. I run both of these ventures with my partner, and we enlist contractors for some operations and software development.

We're a small team, and AI tools are a big help. Last week, I saw a few posts online showing the amazing things that Anthropic's new Fable model can do.

I was keen to try this new technology, but didn't get much of a chance to use it. A few hours in, I was mid-task when the US government forced Anthropic to cut off foreign access to Fable with little to no notice.

The rug got pulled from under me pretty quickly, but because I was well-prepared, it didn't have a hugely disruptive impact on my business. It's a reminder that you can't rely too heavily on AI as a founder, and you should always have a backup plan in case of unforeseen circumstances.

I was keen to give Fable a try, but it was short-lived

I like using OpenAI's Codex for repetitive, code-intensive work, and Claude for tasks that help design the product's aesthetics. AI has been able to completely change the architecture of our codebase in a day, whereas a task like that would've taken a developer weeks to do manually.

After seeing so much about it online, I wanted to use Fable to conduct a full review of our product for safety and security flags. The model was in the middle of making some key changes to our codebase when it got shut off instantly with a notice saying, "Claude Fable 5 is currently unavailable."

I didn't realize until the next day that this had happened because the US government ordered Anthropic to block foreign access to the model. It's been a bit of a bummer, and I feel bad for the people at Anthropic for making a brilliant product and having the rug pulled from under them, too. I'm also quite annoyed we didn't get to do more with Fable. I think it could've propelled us so much further.

Being prepared helped us avoid a huge disruption

This isn't the first time we've had issues with using Claude. In the past, when we used Opus 4.6, it would stop mid-task because it hit the token limit so quickly. We didn't realize how token-heavy the tool was, and it left our codebase in a bit of a mess.

Because we'd learned this lesson with 4.6, we made sure we were more prepared for unforeseen circumstances with using AI.

When we started our product review last week, I asked Fable to create a guide that both Claude or other AI models could follow. This enabled us to pass the remaining tasks to other agents when we lost access to Fable. We passed some to Codex and others to Claude 4.8. If we hadn't been prepared this way, the Fable issue could've resulted in lots of work being out the window.

Fable getting pulled didn't have a major impact because we were ready for it, but it ruined our momentum. We're working on a deadline, and every minute counts, so delays like this can be quite disruptive.

Always have a plan B

This Anthropic incident has solidified my conviction that you can't depend completely on AI.

If the government were to shut off AI access completely, our business wouldn't end, because we've already built out our platform, but we are quite dependent on AI. A situation like that would likely increase our costs, partly because we'd have to switch to the old-school method of hiring developers.

In today's AI era, it's important to always have a plan B. Don't just rely on one AI tool. It's good to understand the strengths of different models.

Make sure you're documenting things as you go by keeping records that exist outside your AI tool. If Claude knows all about our code base, but it gets pulled tomorrow, would I be able to give that over to a developer? At this stage, I think I could, because I've been documenting everything as I go. It's a fail-safe.

A spokesperson from The White House told Business Insider, "The Trump administration is collaborating with AI industry leaders to balance cutting-edge innovation with national security concerns that affect both the United States and our allies."

Anthropic did not immediately respond to requests for comment.

Do you have a similar story to share? If so, you can reach out to one of the reporters at aapplegate@businessinsider.com and ccheong@businessinsider.com.

Last June, fashion models for the fashion retailer Rainbow received a warning: AI was ramping up, and the number of workers needed would be ramping down.

"You may have already seen some changes taking place both within the studio and on the site," wrote Rainbow's studio manager, Phil Caraway. The company had started "styling certain products, and generating avatars, with the assistance of A.I," he explained, and while he couldn't say for certain whether any freelancers would lose their jobs, he wanted them to "plan accordingly."

"Fewer people will be needed in the long term," Caraway wrote in the previously unreported email. "It is very likely that this Fall will see a huge increase in A.I. use."

Thus began what several models described as a year of anxiety and, later, anger. They could see the company using AI to create synthetic models within view of where they worked, the models told Business Insider. At the same time, the models' days in the New York office began to dwindle, they said, leaving many without work. Nearly a year after that June email, Rainbow has begun rehiring some models — though many remain out of work.

In March of this year, the models began noticing Rainbow marketing images that looked like them, but posed in positions or locations that differed from the photo shoots they had participated in. Many suspected the doppelgängers were the result of AI. The lookalike models cropped up across Rainbow's site, social media, and newsletters. A flurry of emails to Rainbow followed, along with a lawsuit by one model.

As AI technologies improve, workplaces across the country are experimenting with how to use them — and navigating the thorny question of their impact on human jobs. Creative industries like modeling are especially exposed as AI-generated photos and videos improve in quality.

AI is growing more common within the fashion industry. In a 2025 study from the Worker Institute at Cornell University ILR School and Data & Society in partnership with the Model Alliance, researchers said that e-commerce gigs were "more vulnerable to displacement by AI technologies."

Business Insider spoke to multiple Rainbow employees and contractors, all of whom requested anonymity, and also reviewed dozens of email exchanges and images, as well as modeling contracts.

"Rainbow is responsibly evaluating emerging AI technologies in the marketplace, and has and is committed to doing so in a proper manner," David Cost, Rainbow's chief digital officer, wrote in a statement to Business Insider.

In a follow-up email, Cost wrote that "Rainbow's dealings with its employees and independent contractors are private" and that the company disagreed with "much of the purported 'facts.'" He declined to comment on specific questions sent by Business Insider. "Rainbow has acted appropriately and in accordance with its commitments, including contracts signed by models," he added.

Here's how Rainbow's AI model experiment got messy, according to its workers — from a slowdown on human modeling work to contract disputes and hiring some of the models back.

Rainbow, founded in Brooklyn over 90 years ago, has over 800 stores nationwide and is privately owned. The retailer caters to thrifty consumers with steep discounts, similar to Fashion Nova or PrettyLittleThing. It also operates the similar brand KissDon'tTell.

For its e-commerce shoots, the Rainbow team looked for models without agency connections, one former stylist who helped recruit models said. Two models said that they were found on Instagram and had little paid modeling experience. Fees varied by model, though many said they made around $50 an hour.

Three models said that one Rainbow employee told them to be available for five days of work a week. The former stylist said that Rainbow asked its freelancers to be available Monday through Friday, but that it wasn't written into their contracts. Two models said they left their prior jobs for the company.

Partway through 2025, the models began to notice something different in the studio: AI training. Employees would lay out the clothes on a flat board, take photos, and upload them to an AI program called Lica, one employee said. Lica generated fully synthetic AI models — not duplicates of human models — for Rainbow, the employee said.

The AI training caused significant anxiety among the models, they said. Trying to lighten the mood, some models said they would crack dark jokes about the system replacing them. Two models said that they recalled instances where the fit of a garment on their body was compared to an AI avatar, pointing out where the avatar needed to be more realistic.

After Carraway's June email a year ago, the models braced for their work to drop off. For months, several models said that they continued to get consistent bookings. Then, they slowed down, the models said, and by mid-March of this year, the work dried up. Some models submitted their availability but said they received no response.

During that period, two Rainbow employees who are not models said that they went weeks without seeing any human models in the studio.

Meanwhile, the models started spotting their doppelgängers on Rainbow's social media.

The models had previously participated in product shots wearing Rainbow apparel, such as a long floral dress, while photographed in front of plain backgrounds.

The doppelgängers they later noticed looked strikingly similar — the same builds, facial features, and outfits they had worn — but were pictured with their bodies in entirely different positions. The models texted these images back and forth in a group chat. Business Insider viewed over a dozen such images.

The second clause of the contracts many of the models had signed allowed Rainbow to use their images "whether intact or in part, composite or distorted in character or form, cropped or altered, without restrictions as to changes or transformations."

One image that sparked conversation in the group chat showed what the models suspected was an AI lookalike that altered the model's original skin tone. The model and the suspected AI lookalike had some similarities — the hairstyle and placement of the hair part, as well as the accessories and shoes — but also some differences, such as the nose shape.

None of the employees Business Insider spoke to had directly seen the creation or editing of these doppelgängers.

Several of the models who suspected that Rainbow was modifying their likenesses with AI raised issues with the company via email.

One of the models, Francheska Pujols, sued Rainbow on May 22, alleging the images defamed her and caused confusion over her endorsement of the company's products, among other allegations.

Pujols wrote in an affidavit that her contract only covered images captured in photo shoots, and "does not in any way authorize the creation of entirely new images, scenes, poses, or compositions that did not exist in the original content."

Rainbow posted photos of what Pujols said is her AI doppelgänger; in one, she straddles a barstool. Another shows her seated, wearing a short skirt, with one leg raised.

Pujols wrote to Business Insider that she would "never pose with my legs open or position myself in a sexualized manner for the world to see."

"I am extremely emotional and have many sleepless nights with the thought of the altered images of me," Pujols wrote. "I sought a professional aide to help with sleep and reconciliation."

Pujols withdrew her suit on May 29 to pursue a private settlement, her attorney wrote in an affidavit. She refiled the lawsuit on Monday.

"As Rainbow has stated previously in relation to this matter, Ms. Pujols' images were used properly and in accordance with the agreement she signed," Joan McGillycuddy, Rainbow's chief legal officer, wrote in a statement to Business Insider. "There is no violation of her rights."

Rainbow's contracts said the models would receive double their day rate for image use outside that second clause. Some models requested compensation for the suspected AI images but were turned down, according to their messages, which were viewed by Business Insider.

Then, the contract back-and-forth began.

On March 10, amid the work slowdown, Caraway sent an email to the models. "To account for today's rapidly-changing technology and expectations of use, Rainbow has come up with an updated Model Release," Caraway wrote.

One clause in the new contract was particularly controversial — one that the models interpreted as granting Rainbow sweeping AI rights.

The new clause allowed Rainbow to use "various technologies, tools, or production methods now known or later developed, including automated or computer-assisted techniques." The clause should be interpreted "broadly" as long as the company was not "materially misrepresenting the model," the contract read.

Some of the models said they refused to sign it. On March 28, Carraway emailed the models that Rainbow agreed to remove a non-compete clause, but the technology usage clause was presented as a dealbreaker.

"Rainbow cannot adjust the AI clause," Caraway wrote. "In order to continue to be hired, this must be agreed to."

It's not clear if the contract negotiations contributed to or prolonged the work slowdown.

Cost, Rainbow's CDO, hyped up the AI program Lica in an April video reposted by the startup's cofounder.

"It's amazing what the people at Lica have been able to do," he said. "We're using them for product photography. We're also using them for editorial or things that you'd see on a homepage or in an email."

Two staffers said the tool was buggy. Some of the synthetic models' legs were too short, one said; the AI repeatedly generated one synthetic model with a white cardigan over her clothes. Creating an AI image would also take long stretches of re-prompting, they said, often around 15-30 minutes.

Rainbow is no longer using Lica, one staffer said. Lica told Business Insider in a statement that it is "focused on foundational AI research for multimodal design models."

"As part of our research efforts, we provided interested enterprise partners with early access to emerging AI capabilities and model technologies," a Lica representative wrote. "We do not direct, supervise, or control our customers' implementation decisions, and we do not publicly comment on specific customer use cases."

Rainbow began bringing some of its human models back at the end of April, employees said.

This time around, some of the models received an agreement with the following clause: "Company will not create digital replicas, train AI on Model likeness, or generate synthetic images not based on original Content."

Rainbow is still producing images of the AI avatars, one staffer said, but not with Lica.

Cost, the company's chief digital officer, referenced the state of AI experimentation at Rainbow in his LinkedIn job description.

"Every experiment designed to replace a person with AI failed," Cost wrote. "Every experiment designed to give a talented person more capability won, and won bigger than expected."

A reação positiva dos mercados no Brasil neste começo de semana diante do fim da Guerra do Irã, com queda do dólar e alta da bolsa, acompanha um movimento global. No curto prazo, a redução dos preços do petróleo vai ajudar o câmbio, reduzir a pressão inflacionária e melhorar as perspectivas para setores mais sensíveis aos juros. Mas, no médio prazo, o cenário não vai ser o mesmo de antes do conflito.

A tendência é que a bolsa retome o rumo das altas com ajuda de setores que se beneficiam desse cenário mais otimista, como varejo, construção civil, shopping centers e ações de companhias de menor porte, as small caps – ainda que o Ibovespa fique pressionado pela correção das ações da Petrobras, que respondem por 12% do índice e, com o fim da guerra, perde 4% nesta segunda-feira (15).

Juros futuros

Os juros futuros também caem, dado que passam a embutir menos prêmio de risco diante da perspectiva de alívio sobre os preços globais de energia. Isso significa que quem aproveitou os momentos mais críticos para aplicar em títulos com taxas elevadas, como IPCA + 8% ao ano ou um prefixado acima de 14% anuais, verá seu saldo em reais na carteira aumentar.

Mas a janela dos juros em patamares historicamente elevados ainda está aberta. O Tesouro IPCA+ 2032, por exemplo, ainda está com taxa acima de 8% ao ano, enquanto os papéis prefixados se mantêm acima de 14% ao ano.

Isso significa que, mesmo com o fim da guerra, a volatilidade dos títulos atrelados à inflação e dos prefixados deve se manter ao longo dos próximos meses. As incertezas continuam elevadas no lado doméstico, com ameaças de novas tarifas dos EUA e diante da aproximação das eleições.

Portanto, o investidor que busca travar os rendimentos deve estar preparado para o sobe-e-desce de taxas e preços de títulos. Por outro lado, quem pode manter os recursos aplicados no longo prazo, no mínimo até 2027, vai se beneficiar do juros nos maiores níveis em vários anos.

Na plataforma Tesouro Direto, os efeitos do acordo entre Estados Unidos e Irã sobre a guerra têm se traduzido em queda expressivas de taxas. O Tesouro Prefixado 2029, por exemplo, saiu de 14,97% no pico da semana passada, visto da quarta-feira (10), para 14,32% nesta segunda-feira (15). Esse recuo de taxas representa um ganho de 1,57% no valor do papel em uma semana.

O Tesouro IPCA+ 2032 também apresentou ganhos no seu valor com o recuo de taxas. O título saiu de um pico de inflação mais 8,36% no momento mais tenso da semana anterior para os atuais IPCA + 8,04%, com tendência de queda. A variação representou uma alta de 2,1% no saldo em reais.

No IPCA+2040, mais longo e, portanto, mais sensível a mudanças de taxa, a alta no saldo é de 6%.

Bolsa na berinda

A questão é o que vem depois desse alívio imediato. Mesmo que a guerra tenha durado pouco menos de quatro meses, o mundo já não é mais o mesmo de antes de fevereiro deste ano, quando eclodiu.

O movimento de diversificação internacional voltou a perder força entre abril e maio. Os investidores globais registraram só em maio vendas líquidas de R$ 13 bilhões na bolsa brasileira.

Ainda que o saldo no ano ainda seja positivo em R$ 43 bilhões, em maio e início de junho mais de R$ 22 bilhões deixaram o mercado acionário local.

E o que isso significa para os investidores? No caso da bolsa, o novo mundo pós-guerra pode manter ventos contrários ao movimento de entrada de recursos que sustentou várias quebra de recordes do Ibovespa no primeiro trimestre.

O momento, portanto, é ajustar estratégias. Uma forma de se precaver de oscilações mais intensas da bolsa daqui para a frente é se tornar mais seletivo na escolha dos papéis.

O setor de “utilities”, por exemplo, eleva o lado defensivo da carteira de renda variável. O nome em inglês se refere às empresas de utilidade pública, como saneamento e energia elétrica. Esse setor inclui nomes como Sabesp, Axia, Copel e Eneva.

São, em geral, negócios mais estáveis e com receitas previsíveis por serem reguladas. Grande parte das companhias também costuma aparecer com frequência nas listas de maiores pagadoras de dividendos.

Setor de tecnologia volta a decolar

Um dos principais ventos contras para a renda variável vem da retomada da “força gravitacional” das ações de tecnologia, em especial aquelas ligadas à cadeia de inteligência artificial. O repique recente ocorre em meio às notícias sobre novas ofertas iniciais de ações trilionárias.

O IPO da SpaceX, do bilionário Elon Musk, movimentou US$ 75 bilhões e levou a fabricante de foguetes e operadora do sistema de internet via satélite Starlink a um valor de mercado de US$ 1,75 trilhão.

A OpenAI, criadora do ChatGPT, deve ser a próxima. A companhia pode captar US$ 60 bilhões em uma abertura de capital no segundo semestre. Esse volume de recursos colocaria a companhia no seleto clube do trilhão, com um “valuation” de US$ 1 trilhão.

Outro mega IPO no horizonte é o da Anthropic, rival da OpenAI e dona da plataforma Claude.

As narrativas de inovação disruptiva — IA, computação avançada, infraestrutura espacial, defesa tecnológica — têm levado boa parte do dinheiro internacional de volta à órbita do mercado americano.

Juros das Treasuries

Outra mudança trazida pela guerra diz respeito ao nível dos juros de longo prazo americanos. Antes do conflito, a Treasury de 10 anos, principal referência nesse mercado, negociava a taxas levemente abaixo de 4% ao ano. No cenário atual, o retorno tem variado entre 4,4% e 4,5% anuais. Na manhã desta segunda feira, com o acordo de cessar-fogo digerido pelo mercado, a taxa operava em leve queda, a 4,45%.

O título mais longo do Tesouro americano, a Treasury de 30 anos, chegou a tocar o maior nível em 19 anos no fim de maio, aos 5,196%. Nesta segunda-feira (15), a taxa permanece perto de 5% ao ano, aos 4,994%. O rendimento atual se mantém no maior nível desde julho de 2007.

Na renda fixa, quando as taxas sobem significa que os preços dos títulos caíram. Com as altas atuais, a remuneração das Treasuries volta a patamares historicamente elevados. O avanço dos retornos passa a atrair mais capital ao mercado dos EUA e fortalece o dólar globalmente.

A alta das taxas vem a reboque das preocupações com a escalada da inflação global. A percepção é que a alta de preços criada pelo conflito não vai desaparecer automaticamente.

Há uma aceleração inflacionária “contratada” e espalhada pelas cadeias globais: fretes marítimos mais caros, seguros elevados, custos energéticos ainda represados e uma normalização da oferta de óleo e gás que ainda deve levar alguns meses. Mesmo com o petróleo em queda, parte desse choque vai contaminar os índices ao longo dos próximos trimestres e tirar a visibilidade futura da inflação.

E isso muda tudo nos mercados. Se antes da guerra os bancos centrais globais ainda administravam ciclos de cortes de juros, agora o caminho se inverteu: o Banco Central Europeu (BCE) subiu a taxa em 0,25 ponto percentual e outros BCs, como o Bando da Inglaterra (BoE) e o Banco do Japão (BoJ), já começam a sinalizar a possibilidade de subir as taxas nas próximas reuniões.

Volatilidade em alta na renda fixa brasileira

No médio prazo, portanto, a renda fixa brasileira pode ver a volatilidade voltar ao mercado de juros prefixados e atrelados à inflação. Se nesta segunda-feira predomina o alívio, daqui a alguns meses, no entanto, tudo pode mudar.

A perspectiva de freio no ciclo de queda de juros aqui e uma guinada lá fora vai afetar as taxas locais. Isso porque, diante das incertezas sobre as decisões dos BCs, os preços vão oscilar ao sabor do noticiário e das preocupações de investidores.

Para o investidor brasileiro, o importante é ter sangue frio. Quem aproveitou para travar taxas historicamente elevadas em títulos IPCA+, por exemplo, tende a ganhar com o chamado “carrego”, ou seja, pelo fato de manter o papel em carteira por um tempo mais prolongado ou até o vencimento.

Há, por exemplo, perspectiva de ganhos se a inflação subir mais do que o esperado. Mas também com uma queda de taxas em um cenário mais favorável. Quem tem mais tolerância ao risco pode, inclusive, aproveitar os momentos de estresse para travar novas taxas.

• After months of uncertainty, a Meta data scientist said she felt a sense of relief upon getting laid off.
• Moyan Chen said the loss of her job made her question what she wanted to do next.
• She's considering AI startups, seeing more risk in traditional data roles at big companies.

Moyan Chen was laid off from her role as a data scientist at Meta in May after just under a year on the job. The 24-year-old, who lives in New York City, isn't sure what she wants to do next. Business Insider has verified her identity and former employment. The following has been edited for brevity and clarity.

When the rumor of layoffs at Meta leaked in March, there was no timeline. Some of my colleagues and I were fearing Wednesdays because Meta has sometimes laid off people on those days. So, every Tuesday night, when I left work, I wondered if I would be coming back.

On Wednesday mornings, I would wake up early to check my email. That lasted for a month, until April, when there was a date for the layoffs: May 20. When the day finally came, and I got laid off, I was like, "This is it." It was more like relief than pain.

I feel like, ultimately, I lost my job to AI.

A lot of my coworkers were also impacted, and they're trying to find jobs. They are making posts on LinkedIn and asking for new opportunities. It feels like we are all sailing on the sea, and Meta is a huge ship that's moving very fast. When the AI storm comes, is your next move to jump to a smaller, slower ship?

Some people I worked with were saying it's better to find a job in finance because it takes longer for them to adopt AI. But ultimately, is the same thing going to happen to you?

A switch in my career path

After I got laid off, I wasn't that nervous, because I'm single and have no family in the US. My parents have been wanting me to go back to China anyway. That's the worst-case scenario because I love the US and the energy of New York City.

I don't know if I plan to find another job at a big company. I have interned at three of them, and now I don't want to climb the corporate ladder. I used to wonder, "How am I going to feed myself if I don't work for a big company?" That's why I didn't resign from Meta. I kept working, and I worked hard.

Now I feel like it's not safe anymore, like I can get laid off at any time. Meta has been very generous with severance, so I have a couple of months to figure out what I'll do next.

I don't think this layoff is a bad thing for me. It's more like a switch in my career path. It's making me see that I could live a different life, and it's probably better than the corporate life.

I'm still in a transition period and don't have all the answers. Seeing how AI is changing things, it makes me rethink the type of job I might want. I've started creating content online to document my career journey and what I'm learning about AI. I'm also interested in exploring career coaching to help people who are experiencing this transition brought about by this new technology.

The longer-term risk

Whatever I end up doing, I expect AI will have an impact. At Meta, I was a data scientist working on Instagram. For that kind of job, the more repetitive tasks are definitely going away. So, writing queries and spending time creating visualizations — these things have already been replaced by AI in Big Tech.

If you only know how to code, that's not enough. If you're just writing SQL queries, using Python, or tracking and analyzing metrics, it's not a very promising career anymore. There will still be a role called "data scientist," but they will need to know more about other functions. There is this emerging trend that requires us to have broader skills and knowledge because of AI.

It got to the point where I wouldn't check AI-generated queries because they have gotten so accurate. I thought that if AI made a mistake on a specific task, I would make 10. For big, ambiguous projects, AI would still make a lot of mistakes, but for specific tasks, it was super accurate. It's very much like a talented individual contributor.

I'm less interested in AI as a stand-alone technology and more interested in how it changes the way people work and build products. If I come across a team that aligns with my interests and values, I would seriously consider joining an AI startup.

Those companies can be risky, but staying at a big company doing traditional data analytics and reporting jobs just feels like I will be left behind. That's riskier in the long term.

Do you have a story to share about your career? Contact this reporter at tparadis@businessinsider.com.

• Many Americans are resisting the construction of massive AI data centers in their communities.
• Some local and state governments are issuing moratoriums on data center development.
• Others have banned data center construction altogether.

Americans across the country are rallying against the AI-inspired data center boom.

Now, an increasing number of local lawmakers are backing them up, issuing restrictions, imposing moratoriums, or outright banning construction.

Data centers have become a major source of contention in the United States, where tech leaders, developers, and investors are pumping billions of dollars into the large-scale construction projects. The facilities house the servers powering the AI products sold by Big Tech companies and leading AI startups like Anthropic and OpenAI.

Although data centers in America aren't new, the AI revolution is fueling ever-increasing demand and requiring facilities that dwarf those of the past. A Business Insider investigation published last week found 1,416 data centers already built or approved for construction across 45 states and Washington, DC, in 2025.

The White House has supported this push for more data centers. In 2025, the Trump administration accelerated federal permitting for their construction and directed the US Secretary of Commerce Howard Lutnick to provide financial support for certain projects. The administration also backed the Stargate project, a joint venture between OpenAI, Oracle, and SoftBank that seeks to build out AI infrastructure.

Supporters say the data centers will create new jobs and strengthen the economy, and are necessary if the United States wants to compete with China to lead the world in AI. Critics, however, are more worried about how they will affect the largely rural communities where they are being built.

They worry about the impact on the environment, wildlife, water resources, air quality, electricity costs, traffic, and noise levels. Some have also criticized local officials and developers for what they say has been a lack of transparency in the approval process. Protesters have swarmed community meetings, launched petitions, and even taken legal action to stop data center developments in recent months.

Some high-profile figures in the data center game, including Jeff Bezos and Kevin O'Leary, have tried to sway public opinion. That PR push, however, hasn't had much impact. A Pew Research Center survey earlier this year found that the more Americans learn about data centers, the worse they feel about them.

• Mark Lanier said he used AI before and during the social media addiction trial earlier this year.
• Lanier, who won the case against Meta and Google, said AI has transformed his workflow.
• He swears by an AI tool that he pays six figures annually for called Boodlebox.

One morning in February, Mark Lanier woke up after four hours of sleep and started preparing to cross-examine one of the wealthiest people in the world: Mark Zuckerberg.

His team had worked through the night, preparing material for the day ahead that he could then review in the hours before court, all with the help of AI.

Lanier, a nationally known Texas trial lawyer with a reputation for taking on major corporations in high-stakes trials, was representing the plaintiff in a landmark social media addiction case. He said AI allowed his team to do significantly more with the limited hours they had to prep outside the courtroom during the trial, which lasted over a month.

"It's as if I have 10 additional workers who are incredibly well-trained, who know the file inside and out, who work 24 hours a day and don't even need to take a break for the restroom, much less PTO," he told Business Insider, adding, "In the 10 hours I might be working outside of court, I can get 30 hours of work done."

AI in law has been touted both as a major opportunity and a cautionary tale, with many stories of hallucinations and fake citations. While the legal industry grapples with how to use AI, Lanier said it's been a "total game changer" for him.

Lanier won the case against Meta and Google, in which the jury found the companies negligent and ruled they knew their platforms were "dangerous" but failed to warn the plaintiff, who was awarded $6 million. The case was a bellwether for thousands of similar lawsuits brought against social media companies.

While Lanier had used the most popular AI products, he said the AI tool he relied on before and during the trial was Boodlebox, calling it "Disney World compared to a swing set in the backyard."

A leader in the education technology space, Boodlebox provides access to major models like ChatGPT, Claude, and Gemini, allowing users to switch between them or compare results. It's also collaborative, allowing Lanier and his team of lawyers to work with the AIs in the same digital workspace.

Lanier worked with Boodlebox to create a custom license that costs him six figures annually and is tailored to his needs.

"We could, in essence, take my brain, take 42 years of my experience, take the things that I have learned and studied and published and not published and incorporate it into the brain that drove my AI queries and results," he said.

He relied on AI before and during the landmark trial

Lanier is careful when talking specifics about how he deploys his AI. He says it's a matter of "trade craft" and that his firm is "doing some things that nobody else is doing."

One example he gave included taking transcripts from court each day and asking different models to evaluate them. He said AI is also great for finding a more creative or visceral way to describe something in court. He even would feed AI jury notes that came up during deliberations and ask it to evaluate where the jury was in the process.

At the end of court each day, they'd meet in his war room, debrief, and assign tasks to everyone, such as pulling the five most critical documents supporting point A. The team would then break and do much of that work in Boodlebox, allowing him to review what they've put together and how. He said he and his team, which includes several of his daughters, spent thousands of hours on the platform.

While most of Boodlebox's clients are big universities, a company representative told Business Insider that the platform is also exploring more enterprise and law adoption, in part because of its work with Lanier.

Lanier said he doesn't use AI in the way that often gets people into trouble. "I'm not going to say, 'Go do my research and write my brief,'" he said, adding that there was one instance in the case where AI cited something from the record and he knew it wasn't correct.

"It's not unbridled," he said. "You are an important part of the equation."

His advice to other lawyers trying to use AI was to keep up with the developments in the rapidly evolving field. He has an AI team at his firm that sends him a document every Friday with all the developments in AI, typically three pages single-spaced.

"Next trial, I will make what I did last trial look like Fred Flintstone and the Stone Age," he said.

Brian Rezendes anticipates his retirement years will be filled with AI agents, algorithms, and APIs — along with the occasional vacation with his wife.

Rezendes, a former pool business owner, retired in April from a retail job in rural North Dakota. Like many retirees, the 64-year-old envisioned his post-work years as a time to relax, travel, and stay active. He did not expect to be neck-deep in conversations with chatbots, vibe coding websites, or building YouTube channels. Though he'd always been interested in technology, he rarely delved into the deeper plumbing behind it until a few years ago, when he became immersed in AI. Nowadays, he spends almost all of his time building apps… until the real world comes calling.

"My wife gets a little bit jealous when I spend too much time on the computer," Rezendes says.

Retirement has gone digital. In recent interviews, 15 retired Americans admitted they and their friends are glued to their screens, perhaps to a fault. Hours they could have spent tidying up the house went toward learning the best AI tools and, as three tech-savvy baby boomers put it, "staying current." Some post-career Americans who moved abroad said tech is all the rage in their beachfront expat communities. Retirement communities have swapped watercoloring for AI education. Starting an AI-powered business replaced the golf course. ChatGPT is the new nurse's assistant. Robots are some older Americans' new best friends.

Dee Humphrey is among them. The 73-year-old in Schenectady, New York, has used a companion robot called ElliQ for over three years. And while she's waiting for a new version to arrive, she's been having "withdrawals because I can't do anything with her."

The new reality of retirement isn't all screen addiction. Some of this development has been a boon for older people navigating a new phase of life. In Austin, Edward Perry, 72, said that he used AI after a terminal cancer diagnosis to "help me with living as rich and full a life in what time I have," including managing his health and finding ways to be more present in his family's lives.

"As I'm getting older, I have more aches and pains, but with utilizing these new technologies, I'm going to be able to do more and more," Rezendes says.

Many others acknowledged the risks of getting too hooked on tech. Most knew that relying too heavily on AI meant losing agency and receiving potentially faulty information. Others said being too invested in tech could mean less time staying active. Some noted that after decades of work, these were their years to relax, but they couldn't bring themselves to close their MacBooks.

If Gen Z is the first generation to grow up on the internet, baby boomers are learning how to be the first generation to retire on it.

Unexpected and omnipresent

For those in retirement, screen time of all types has been increasing. Surveys show that adults 65 and over almost doubled their YouTube consumption on TV from 2023 to 2025, and older Americans spend over four hours a day in front of screens. Brittne Kakulla, senior research advisor for AARP Research, says the group's Tech Trends survey found smartphone ownership among adults aged 50-plus skyrocketed from 55% in 2016 to 90% in 2025. Perhaps more striking was the number of older people trying out AI. Use nearly doubled from 2024 to 2025, from 18% to 30%, and many more said they are interested in experimenting.

Nearly all older tech superusers I spoke to were surprised by the amount the tools had become integrated into their retirements. Jan Friedlander, 81, used online databases in her real estate career, but only became hooked on tech a few years ago after she left her job. As she battled cancer and macular degeneration, she used AI to guide her treatment, and soon found herself relying on it to research clothing, plan vacations, and more. As she became more prompt-savvy, she felt confident enough to start teaching her peers.

"I've always had a curiosity about things that would come along that were new," Friedlander says.

She also began facilitating AI classes in Denver for those 50 and over with her friend Pat Smith, 73. Smith, who has a more technical background in consulting and pharmaceuticals, says the classes have attracted many "eager retiree students." Smith also sees both sides of the AI boom. On the positive side, she submitted her lab work to ChatGPT after having a reaction to an antibiotic, prompting her to follow up with her doctor and allergist. But she also bemoans the disappearance of human customer service and the online portalization of medical care. To combat the AI creep, Smith has monitored her tech usage, maintained a regular exercise schedule, and worked on mosaics.

"I have friends who are losing their mobility, moving into assisted living, and have gotten terminal diagnoses, and I know that's all around the corner," Smith says. "I'm hoping I get some more time to do what I've been enjoying the last few years."

Working with tech

While cutting-edge tools have become a retirement fascination for some, many older Americans are unexpectedly working into their later years and, by extension, learning new tech tricks. For my 80 Over 80 series, I spoke with dozens of workers in their 80s, many of whom couldn't afford to retire and now had jobs that required AI. At 72, Marcia Sweet's home is fully synced with robot vacuums and smart lights, and she runs a tech support business in Bradenton, Florida. She can't afford to stop working, as the extra money goes toward financing her eventual long-term care, and she hopes AI can supercharge her business.

"I'm still like a little kid with a toy about technology, with the same kind of excitement," Sweet says. "I'm kind of addicted."

Other older workers used tech to pivot later in their careers. A decade ago, Laura Noren, now 61, was weary of her career as a registered nurse, so she opted for an unexpected route — IT classes at a local college in Michigan. The learning curve was massive, as most of her 18-year-old classmates grew up steeped in tech. She later supplemented these classes with online courses on programming languages and databases.

"I envisioned myself retiring at 60 and no later than 62. My husband and I would be fully retired and never work again, moving into a condo and doing plenty of traveling," Noren says. Instead, "he left his job earlier than planned as a corrections officer, and I was managed out of my company. We had to change our plans."

The courses didn't necessarily prepare her for her current job as an Amazon Flex driver, which gives her the flexibility to care for her "technophobic" 84-year-old mother with memory issues. But her skills have come in handy when teaching her mother how to add phone contacts to favorites or avoid scams, and Noren hopes to find work down the line that better suits her skills. She still hopes to have some version of the retirement she envisioned years ago, but expects tech to play a bigger role.

Others who returned to school in their later years said they've integrated age tech into their lives for peace of mind. When Mark Bayer, 63, decided to retire from his community banking career at 60, he thought, "I will never have to sit through another damn Zoom meeting again, and I'll be the happiest person in the world." To his surprise, he began teaching English as a second language over Zoom and reenrolled in college to be "exposed to new ideas from younger minds." Bayer, who lives in Pennsylvania, expected his classmates to debate and brainstorm ideas off the top of their heads, but they all went to ChatGPT instead. Initially, he was dumbfounded. But when he saw the list of ideas for a group discussion, it exceeded what he would've come up with.

Ignoring AI, he says he realized, "is a way to say I'm done learning anything new, which is self-limiting."

There have been downsides: He's noticed that disconnecting from tech has become harder. He admits that if he gets a call while mowing the lawn, he will stop to pick it up. His wife is the same way, sometimes scrolling Instagram for hours without noticing. He hasn't quite erased the idea that face-to-face interaction has some merit, though.

A robot-enabled retirement

Many new high-tech tools are being built to help older Americans remain healthier and safer in their homes and assisted living communities. Chia-Lin Simmons, CEO of medical alert devices company LogicMark, tells me that technology in caregiving has become a necessity rather than a luxury, with the potential to predict falls and detect Alzheimer's early. AI is being trained to track behavioral patterns and health outcomes, though it sometimes falls short at triaging calls and often erases the human element, isolating older Americans who need the company most.

Some boomers are ready for this Jetsons-like future. Take Michelle Murphy, 64, who is pursuing an MBA with a concentration in AI. A photographer and instructional designer in Michigan, Murphy says her focus in her 60s has been pivoting to a new career— retirement isn't a good fit, she says. Down the line, she isn't opposed to using robotic healthcare workers to avoid assisted care, though she's keen on not becoming overly reliant on tech due to privacy concerns. For now, her goal is to get her coffee pot to start automatically.

"If there's an automation that can help me do the things I need to do, mow the grass for me, pick up heavy things, whatever it is, I'm totally on board with that," Murphy says.

There is a big market in making the idea of robot-assisted care a reality. Investment in age tech has boomed, particularly in products that make caregiving easier, like smart home automation devices, companion robots, and motion sensors. AARP predicts that by 2030, the age-tech market will be worth $120 billion. And given the rise, many hope age tech can alleviate some of the burden for younger generations.

"We've got 63 million family caregivers, 70% of them in paid jobs, and we're very familiar with childcare, but elder care is not well understood," said Diane Ty, managing director of the Milken Institute Future of Aging. "That's what's breaking the backs of so many workers right now."

Plenty of people and investors I spoke to also hope AI and other age tech can slow cognitive decline. However, various studies have shown that AI assistants contribute to reduced cognitive engagement and skill atrophy, meaning in some ways, relying too much on AI works counter to what these super-users may think.

80 is the new 25

As I wrote last year, America's octogenarians have been embracing tech in surprising ways. Frank Engelman, 82, has created apps, runs a YouTube channel, and writes a Substack about tech education. Luis Bautista, 82, told me he was using AI to write a book and start a business that he one day wants to pitch to Y Combinator. Phyllis Scalettar, 80, began an AI education and consulting firm. Karen Shapiro, 80, said this month that she uses AI for everything from planning vacations to Italy to managing finances — "tech will make life less confining and more enjoyable as we age," she says.

Study after study shows loneliness continues to grow among older Americans. According to AARP, 40% reported feeling lonely last year, up from 35% in 2018. Tech may be partly to blame, as an increasing number of older Americans are addicted to their phones — one survey found that 40% of the over 2,000 respondents ages 59 to 77 felt discomfort when pulled away from their devices.

For a lot of Americans, however, tech is a way to make the most of their golden years and to stay healthy for longer.

Marvin Honig, 88, takes AI courses, set up NotebookLM files for his St. Petersburg, Florida, condominium board, and use advanced tech to manage trust accounts for former law clients. Perhaps this could've been expected from an early tech adopter who received tech support from a young Michael Dell. Still, seeing many of his neighbors using all sorts of tech was perhaps not on his bingo card, and many of his interactions now revolve around tech recommendations and support. Like many older techies, the tech wave has also allowed him to luxuriate in the disconnected part of his life, from visiting museums and restaurants to attending in-person community events — he gets there using his Tesla's self-driving feature.

• A group of states is now investigating OpenAI.
• New York State Attorney General Letitia James sent a subpoena to the company on Friday.
• OpenAI said it's taking the concerns "seriously" and is "committed to learning."

OpenAI said it's "committed to learning" after a coalition of states launched an investigation into how the tech startup's products impact users.

An OpenAI spokesperson said in a statement that it's taking the states' concerns "seriously" and will "engage constructively with their offices."

"Today's ChatGPT includes a more protective experience for minors and people experiencing difficult situations, with safeguards that direct them to real-world resources and trusted human contacts," the spokesperson said.

"None of this changes what families have gone through, but we are committed to learning, improving, and getting this right," they added.

New York State Attorney General Letitia James served OpenAI a subpoena on Friday seeking a wide range of documents, The Wall Street Journal first reported. The documents pertained to user engagement and retention, the company's handling of health and consumer data, deep learning models, activities related to young and older users, and more, the Journal reported.

Friday's subpoena is the latest legal headache for OpenAI. The company has navigated copyright infringement claims, privacy lawsuits, and a high-profile trial pitting SpaceX CEO Elon Musk against OpenAI CEO Sam Altman.

Perhaps most concerning, however, are a handful of lawsuits that say ChatGPT contributed to decisions by users to die by suicide. In response to a May report by The New York Times, the company said ChatGPT "is not a substitute for medical or mental health care, and we have continued to strengthen how it responds in sensitive and acute situations with input from mental health experts."

The family of a victim in a fatal campus shooting at Florida State University in April, meanwhile, has also filed a lawsuit against OpenAI. The family says ChatGPT's guardrails failed to recognize the threat in the shooter's conversations with the chatbot. In June, Florida Attorney General James Uthmeier filed yet another lawsuit against OpenAI and Altman over the shooting.

In that complaint, Ulthmeier says ChatGPT has "aided and abetted deadly rampages" and "encouraged vulnerable people into suicide." The state attorney general also says users have become addicted to ChatGPT, a tool that "feigns human compassion to collect their data with no parental oversight."

In response, OpenAI again said it has introduced further safety measures into its products. "Losing a child is the most devastating tragedy that can happen to a family and we know that no words can come close to addressing the pain of such a loss," the company said in a statement at the time.

The coordinated investigation launched by the state attorneys on Friday mirrors a similar investigation into TikTok, which resulted in a 14-state lawsuit now making its way through the courts.

Like the OpenAI investigation, the TikTok lawsuit is led by the attorneys general of California and New York. The states say TikTok knowingly uses addictive features to lure kids, which negatively impacts their mental health.

Lawyers told Business Insider that it's a common strategy for states to band together when they go after multibillion-dollar companies because they are more expensive for the companies to defend and, should a case falter in one state, the suit can continue in another.

It's also how the government went after Purdue Pharma, the maker of OxyContin, in 2017, and the tobacco industry in the 1990s.

• Founders and workplace experts said that post-pandemic AI startups have a different work dynamic.
• AI startup employees often voluntarily come to the office and work longer hours without an RTO.
• Founders said that in-person work fosters a high-trust environment that spurs innovation.

"What is an RTO?"

That was Together AI CEO Vipul Ved Prakash's response when asked by Business Insider whether he had ever had to send a return-to-work (RTO) memo to push employees back to the office of the cloud compute startup.

"People generally like to come in," said Prakash. "We've never enforced it."

Prakash's response illustrates a stark cultural difference between AI startups formed after the COVID lockdowns and long-established corporations, with people voluntarily coming to the office, sometimes on weekends.

Nicholas Bloom, an economics professor at Stanford University, told Business Insider that the age demographics and personal stake many startup employees have in their companies created a work mode that is "almost entirely in-person" and "100% work focused."

"For a single 23-year-old with equity worth $20 million, it makes sense to work in the office for 100 hours a week," said Bloom. "They don't work from home, they home from work."

The tight-knit culture of AI startups

Arvind Jain, founder and CEO of Glean, an enterprise AI for productivity, said he "was not eager" to bring his team members back to the office because finding an office is a hassle, but everyone wanted to be in person and return to their original mode of working when the company first started right before the pandemic in 2019.

"We just simply didn't know how to work from home because everybody was in this one small room," said Jain of the early days of the pandemic lockdown. "We used to be sitting next to each other, brainstorming what to build, and so we found that very, very hard."

Over time, said Jain, he learned to enjoy remote work and got to spend time with his family, but the team genuinely wanted to be together again.

"That's the difference — there's this startup spirit, and it's only 10, 15 people, and we want to be with each other," said Jain. "They love each other, they bond with each other, we used to play games together, and we have very fond pre-pandemic memories as a close-knit group."

Jain said that as Glean grew more rapidly in recent years, it has since moved into a larger office space and dedicated Thursday as its work-from-home day.

Spiros Xanthos, founder and CEO of Resolve AI, an enterprise technology startup that builds multi-agent AI systems, said the company has a "very strong culture" of in-person work and has never had to ask anyone to be in the office.

"We have a fairly big office now, and we have breakfast, lunch, and dinner," said Xanthos. "Most people have lunch in the office together with their colleagues, and many people stay to have dinner in the office."

Xanthos said that since founding the company in early 2024, "cohesion and culture and friendship" among employees has been critical for the company, and that he often brings colleagues based in New York to the Bay Area for offsite retreats so the team could get to know each other better.

"People will actively avoid working remotely at this point," Xanthos added. "Especially for some of the younger folks who didn't have many years of experience, but maybe worked remotely before this, many of them tell me it's day and night — the fact that they have so many friends at work now that they can trust."

AI's innovative nature demands in-person interactions

Richard Florida, an urban studies theorist and professor at the University of Toronto, said the AI wave has unique characteristics compared to other startup booms, which may generate greater in-person demand.

"Innovators have to be close to end users because end users are a part of the innovation system," said Florida, of why it's easier to work in person in the AI industry.

"If you're an AI company, the technology itself is interesting, and you can invent it, but what you really learn is by interaction with the end user, by interacting with your customers and clients," Florida added.

Xanthos said the demand for in-person attendance ultimately boils down to the nature of an innovative industry.

"As a company, we're solving very, very hard problems, and to solve these problems, you operate at the frontier," said Xanthos. "And this means that you need to experiment a lot, try a lot of things that might fail."

"That in turn requires a very high trust in an environment of psychological safety where people feel that they have the ability to innovate bottom up," Xanthos added, "Where they don't need to be told what to do, where there is communication velocity and bandwidth."

So the next time you speak to an AI startup founder, don't ask how their RTO is going — they're probably too busy trying to squeeze everyone into the office.

• The majority of competitive House districts have data centers either planned or under construction.
• Neither party knows how to handle their political fallout heading into the midterms.
• Data centers are shaping up to be a yearslong political slog for both parties.

More than 200 data centers are going up in dozens of competitive House districts — and neither party knows how to handle their political fallout heading into the midterms.

The energy-hungry computing infrastructure being built to meet the explosive demand for artificial intelligence has sparked opposition to rising electric bills, water consumption, use of farmland and influence of the tech industry. That stew of frustration has made data centers the target of campaign ads and a populist fervor that's toppled local elected leaders.

It has also become a rogue element in the races that will decide which party controls the House: The majority of competitive districts — 40 out of 69 — have data centers either planned or under construction, according to an analysis of Data Center Map data by POLITICO, which like Business Insider, is part of the Axel Springer Global Reporters Network.

Even though Republicans represent most of those competitive districts, data centers are shaping up to be a yearslong political slog for both parties. Some 1,500 of them are planned or being built in 232 congressional districts, with a nearly even partisan split. Interviews with and statements from more than 20 congressional candidates, political strategists, and activists make clear that while individual campaigns are trying to shape their positions, broader party messaging is essentially nonexistent.

"There's more political signs against AI in our region than for candidates in the upcoming races," said Democratic Rep. Marcy Kaptur during a hearing this spring. Kaptur is fighting to keep her seat in Ohio's 9th District, where Aligned Data Centers is building a data center that would be used for AI, cloud computing and more. "The public opposition that is arising, it's spontaneous combustion coming up from the grassroots."

The industry's exponential growth means that lawmakers from all parts of the country are now exposed to it, from the dense data center developments in the Virginia suburbs to the heart of the industrial Midwest.

Lawmakers are taking scattershot approaches that range from opposing data centers altogether to embracing them in the name of economic development and national security. Some are eschewing the issue as a local matter, while the White House and Congress grapple with how to regulate the data center buildout.

The White House announced a non-binding agreement in March with technology executives who pledged that their companies would provide their own power for data centers as a way of limiting the economic blow to everyday consumers. Lawmakers have also introduced a handful of bills with similar objectives, from GOP Missouri Sen. Josh Hawley's GRID Act to Democratic Virginia Rep. Suhas Subramanyam's Data Infrastructure Risk Reduction Act to a plan by progressives Sen. Bernie Sanders of Vermont and Rep. Alexandria Ocasio-Cortez of New York to enact a federal moratorium on data center construction.

"People should not want their member of Congress deciding local zoning decisions," said Rep. Tom Barrett, a Republican whose Michigan district both parties' congressional arms are targeting — and where there are six data centers operating and six more planned. "It would be a dangerous precedent."

Data center proposals have spurred marathon city council meetings, scores of bills in state legislatures and ballot measure campaigns to ban their construction in California, Michigan, Nevada, Wisconsin, Maryland, Utah and Ohio, where residents are pushing to put a constitutional amendment on the ballot in November.

"There's not one big national message on this specific thing," said one Democratic strategist working on congressional races, who was granted anonymity to speak candidly. "But in certain districts, data centers are going to be a major, major player."

Asked about its strategy on data centers, Democratic Congressional Campaign Committee spokesperson Viet Shelton said, "While House Republicans fall in line behind failed policies that spike prices out of control, Democratic candidates and Frontliners are fighting for common sense solutions to provide meaningful price relief, encourage economic growth, and meet the unique needs of their communities." The National Republican Congressional Committee declined to comment for this report.

Although they're motivating politics more than before, the number of data centers has been steadily growing for decades. They have spread in lockstep with the growth of the internet, and more than 2,500 U.S. data center facilities are operating across 373 congressional districts, according to POLITICO's analysis. Virginia, Texas, and California contain the greatest number of data centers, and more than one in three Americans live within 5 miles of one that's already operating. In five states, most residents live within 5 miles of one.

Investors plan to spend hundreds of billions of dollars building new data centers, which would push utilities across the country to make massive electric grid upgrades to feed the facilities — expenses that can be passed to everyone who consumes power. Massive "hyperscale" facilities that are owned by major tech companies demand the most power. And though hyperscalers make up a relatively small portion of the facilities now operating, the number in development would increase their count by 74%.

Despite the demand for data centers, there are examples of proposals that are falling through due to community opposition or shifting business calculations. Just the announcement of a data center can be enough to pressure elected officials to act.

In Wisconsin, for example, four proposals have been canceled and one paused following local pushback, according to Healthy Climate Wisconsin, a nonpartisan public health nonprofit whose work includes raising awareness of data centers' environmental health risks.

"We've been hearing from policymakers across the state that data centers are the top issue they're hearing from their community, as far as concerns," said Abby Novinska-Lois, the organization's executive director. "Data centers will definitely be a factor in upcoming races in Wisconsin, and I would say they're a factor already for those who are holding office in their decisionmaking."

How candidates are reacting

Among 69 House districts expected to be competitive, nearly all already have at least one data center, and most have more on the way.

POLITICO asked the 10 House members in battleground districts with the most upcoming data centers what their stance was on data center regulations. Five of eight Republican incumbents, and one of two Democrats, responded.

Their answers illustrate an awareness that voters are in no mood to greenlight anything that will send electricity bills higher. Even the incumbents most supportive of data centers caveat their support with the need to protect consumers.

Iowa Rep. Zach Nunn — who represents a district with 31 data centers planned and 33 already operating, more data centers than any other Republican incumbent in a competitive race — said in a statement that his state is a "model for how workforce development and AI leadership can work hand in hand."

"But I also hear from Iowans who don't want higher utility bills or sweetheart deals for out-of-state tech companies," he said. "And they're right to be cautious."

The tech lobby has shaped up to be a key player in the midterm races. Candidates who are too critical — particularly incumbents — run the risk of losing support from the tech lobby or attracting fierce opposition.

"They're between a rock and a hard place," said Texas-based GOP consultant Brendan Steinhauser, whose clients have included Sen. John Cornyn and Rep. Dan Crenshaw. "Politically, it's not a very smart move to come out and be seen as too close to big tech or doing the bidding of Big Tech, but a lot of the money is flying to them through that."

The advertising so far this election cycle backs that up. All of the congressional and gubernatorial ads that mention data centers, as identified by the political advertising tracker AdImpact,are critical of the facilities. Most attack Republicans for supporting them.

Over an image of cables running from computer equipment, one such ad from the progressive Priorities USA PAC says: "Driven by higher demand for electricity from AI data centers, residents can expect to see a 3% increase in their electric bill. But Pennsylvania's Representative Scott Perry somehow believes we're winning the war on high prices."

Perry, a Republican, told POLITICO he does not support data centers in his district, which includes the cities of Harrisburg and York.

"I don't think it's the best place for it, quite honestly," he said. "Pennsylvania's got a lot of energy in the ground, and the data center to me should be right at this point of energy production and generation, which is kind of in the more rural parts."

Ads from Democrats, including Maryland Gov. Wes Moore, highlight their own records of regulating data centers. Democrats scored some early wins on energy affordability messaging last November as their candidates soared to victory in Virginia and Georgia, promising to place guardrails on data center growth and ensure they pay their share of power costs.

The League of Conservation Voters, an environmental group, launched similar ads targeting Virginia's state legislature incumbents and their data center interests in 2025, and those races were later won by candidates who positioned themselves against data centers. Sara Schreiber, the group's senior vice president for campaigns, said such advertisements were a "powerful" tool for driving home the connection between data centers and affordability.

"There is continuing concern around folks' rising electricity costs," Schreiber said. "They want to support candidates who are showing that they understand, they want to fight against it and have a plan to do so."

Still, Democrats up and down the ticket are open to their construction.

Pennsylvania Gov. Josh Shapiro announced requirements on May 27 for data center developments — including a plan for covering energy costs — but not a moratorium. Paige Cognetti, the mayor of Scranton, Pennsylvania, who is running to unseat Republican Rep. Rob Bresnahan, has said "we are ready for development," but that data centers should be on former industrial sites and not areas that could be used for housing or parks.

The Data Center Coalition, an industry association, has been supportive of the White House's "ratepayer protection pledge" and other policies that require data centers to cover more utility costs. The coalition's senior director of federal affairs, Cy McNeill, said that they're working to educate representatives as constituents increasingly voice concerns about water usage and energy prices.

"How do we provide the facts to the office, or to the congresswoman or congressman, to kind of help educate constituents on this?" McNeill said. "If we actually take a step back, look at the facts, I think the story is a lot different."

Environmental activists say consumer protections don't go far enough. While elected officials who are open to data center construction often focus on keeping energy costs down, activists are worried about other potential risks, too, including water quality and air pollution.

One of the most prominent disputes is out of Memphis, Tennessee, where residents are fighting Elon Musk's expanding xAI data center, the Colossus supercomputer. KeShaun Pearson, the executive director of Memphis Community Against Pollution, said federal action such as the ratepayer protection pledge still allows developers to use polluting energy sources without acknowledging environmental impacts.

The xAI facility, for example, burns enough methane gas to power 280,000 homes — motivating the Memphis organization to take the monitoring of pollution and health risks into their own hands as they push elected officials to act.

"We surely don't want data centers that are directly causing us health issues," Pearson said. "I think our politicians have to understand that and have to move accordingly."

Pavan Acharya and Sean McMinn contributed to this report.

Methodology

The data center locations used for this analysis were based on a combination of Data Center Map, geocod.io and public sources. POLITICO used automation and manual reviews to verify exact coordinates and district assignments, checking against U.S. Census Bureau files.

Data Center Map's data is as of April 30. Data centers included as "upcoming" are those that DCM labels as planned or under construction. The dataset is not an exhaustive list; it is based on voluntary data submissions and collections from providers or other sources. Government-owned data centers are not included.

Some data center companies operate within the same colocation building and lease space to other companies; those cases count as one facility. Facilities that are a part of a campus or multi-tenant building count individually. Cases where the exact facilities within a campus or multi-tenant building are unknown are counted as one data center.

Some data center facility addresses are approximate. In those cases, provided ZIP codes are used to determine congressional districts. In cases where a ZIP code overlapped with more than one district, or if no location information is disclosed for the facility, data centers are excluded from the district-level analysis. The population living within a five mile radius of a data center is determined using only facilities with exact addresses or intersections.

The number of data centers in midterm races account for newly redrawn boundaries finalized in Alabama, California, Florida, Louisiana, Missouri, North Carolina, Ohio, Tennessee, Texas and Utah. Competitive races are based on targets from the Democratic Congressional Campaign Committee and the National Republican Congressional Committee, as well as POLITICO's own reporting.

The Axel Springer Global Reporters Network harnesses the resources of the company's newsrooms to publish ambitious scoops, investigations, interviews, opinion pieces and analysis. It allows journalists — including those from POLITICO, Business Insider, WELT, BILD, Onet and Fakt — to collaborate on major stories for an international audience of hundreds of millions across platforms: online, print, TV and audio.

A SpaceX deu a largada ao que deve ser um período de grandes IPOs nos Estados Unidos, gerando uma nova leva de bilionários aos rankings da Bloomberg e da Forbes. Um levantamento feito pela consultoria Sacra para o jornal The New York Times (NYT) estima que as megalistagens de SpaceX, Anthropic e OpenAI devem criar […]

O post OpenAI, Anthropic e SpaceX viram “fábrica de fortunas”: 20 bilionários e 16 mil milionários a caminho apareceu primeiro em NeoFeed.

• Higgsfield AI made a sci-fi action thriller called "Hell Grind" starring AI actors.
• The movie cost around $500,000 to produce.
• I went to a screening to see how it held up.

For a brief moment toward the midpoint of the AI-generated film "Hell Grind," I caught myself experiencing something unexpected: genuine emotion.

As the male lead, Roco, gazed at a photo of his recently kidnapped love interest, he flashed back to memories of them growing up together in an orphanage. The sadness and yearning felt real.

The sensation didn't last.

Mid-flashback, Roco and his AI-generated costars began laughing in an eerily synchronized fashion, their eyes peeled wide open. As I sat in New York's Metro Private Cinema this week, scooping up handfuls of popcorn, the uncanny valley of AI came roaring back.

Generative AI has crept into a variety of corners of the entertainment business this year, spooking many creatives who worry what it could mean for their jobs. While post-production teams are turning to the technology for de-aging and other effects, some actors in short dramas are already losing out on roles to AI characters. The shift is a top concern for the actors' union SAG-AFTRA, which approved new contract language this week that pushes producers to bargain over the use of synthetic performers.

"Hell Grind," which takes AI usage to the max, sprang up in May at the Marché du Film in Cannes (a side event that's not the famous Cannes Film Festival). The brainchild of startup Higgsfield AI — which runs an AI platform for creatives, brands, and marketers — it was conceived as a way to show the tech's potential as more than just a tool for making short videos. The company, which crossed a $1 billion valuation earlier this year, spent around $500,000 to produce its 95-minute film, with much of its budget going to computing costs. While AI regularly shows up in bits and pieces of Hollywood productions, "Hell Grind" is the highest-profile film made entirely with AI-generated visuals.

Higgsfield tapped a group of in-house creatives and outside filmmakers who used highly specific text prompts (typically around 3,000 words) to generate around 100 hours of content, which was edited down. The company did not use AI to write the script, except for a few short filler moments, which Higgsfield's CEO, Alex Mashrabov, told me he thought were noticeably less effective in the film.

The result is a visually impressive movie with a passable plot line, landing somewhere between a video game and an effects-heavy project like "Planet of the Apes."

"It's a new workflow, and it's also very important for us so that we show to the world what's possible," Mashrabov told viewers at the screening this week. "The production process looks different where it's actually possible to go back and iterate with AI and deliver exactly the emotion which the creative director was envisioning."

The company is releasing portions of the film on YouTube and plans to open source its workflow, production process, and prompts in the coming weeks.

At various points during "Hell Grind," I was taken out of the story when a character did something that just felt … off. The way Roco held a slice of pizza in one scene looked like it was his first time encountering the food, for example. The synthetic children in the movie generally creeped me out, and the AI-generated voice work didn't always feel consistent (one character seemed to flip between a British and American accent, for instance).

Still, it was hard to shake off the feeling that talented AI prompters may soon be coveted players in Hollywood.

While I wouldn't expect to see AI actors or writers playing a big role in the making of films like "Tár" or "One Battle After Another," it feels like this technology will be hard to resist for budget-sensitive executives angling to speed up movie production. That's especially true in genres like action and sci-fi, where visual effects budgets can be a big constraint. And the technology may open doors for independent filmmakers who have grand ideas but small budgets.

"Budgets and opportunities are not equally distributed across the world," Mashrabov said. "Hopefully, this will spark the next generation of creativity."

• Facebook Marketplace sellers are using images of women to draw attention to items for sale.
• There's just one thing: The women aren't real. They're AI-generated.
• I talked to one seller who said the tactic got him a lot of views, but not necessarily more sales.

Let's say you've got a 2013 Jeep Wrangler you want to sell. It's a great car — low mileage, good condition, and a custom-lifted suspension. But there's a lot of Jeep Wranglers out there on Facebook Marketplace — how do you make your listing stand out?

For Rogelio Llamas, it was a little sex appeal that was missing. He added an image of a woman in a bikini top, denim shorts, and cowboy boots leaning on the hood of his Wrangler. Now there's a way to draw eyeballs. Just one thing: The woman wasn't real. Llamas generated her with AI.

Llamas, who's in Southern California, is part of a trend I've seen on Facebook Marketplace lately. He told me he got the idea from a YouTuber who makes videos about how to buy items at thrift stores and resell them for a profit.

On Reddit and social media, people have started to notice a strange new trend: scantily clad AI-generated women propped up against everything from cars to dump trucks to other items for sale.

In one listing for a motorcycle, a hot goth girl appears to be sitting on it. In another, three women in nearly identical outfits (yet slightly too tall) lounge across a 2010 Mercedes. In a listing for what appears to be a dilapidated hot tub, three young women in tank tops and short shorts sit atop the filthy plastic.

On X, someone found a listing for some kind of Caterpillar heavy machinery equipment — featuring a sexy woman in a bathing suit posing on the rusted metal. All had hallmarks of AI images — strange figures, slightly off sizes, and generally improbable looks.

Using attractive women to appeal to car or boat buyers isn't a new concept by any means. Sex sells! But the addition of AI-generated babes to Facebook Marketplace listings feels new.

For one thing, images on Facebook Marketplace typically look kind of crummy. The whole point is you're getting a listing of a real used item sold by a normal person, not a professional. As a frequent Marketplace buyer myself, I'm generally wary of any listing that looks too professional or whose photos are too good — to me, that's a sign the seller knows what they have and may be pricing it too high.

Llamas said that in the influencer video he watched, it suggested using a larger body size for the AI-generated woman rather than a slim one, because it creates a curiosity gap that makes someone want to click. (Who knows if that's the case?) The fact that it seems obviously AI is intentional. "It's supposed to be funny, but also be like, damn, is she real??" Llamas told me over Messenger.

The WTF-factor seems to be a trend. In one Marketplace listing someone posted on Reddit, a seller put an AI sumo wrestler standing on the drawers of a large toolbox, presumably to demonstrate the strength of the drawers and also to catch attention.

But the eye-catching tactic doesn't always translate to a sale. Llamas said that, indeed, he got way more views and clicks on his Jeep listing after adding the photo with the woman. Unfortunately, they're not all serious buyers. "They are messaging if she comes with the car, if she's real, or they think I'm her." He said. "I get negative messages like 'go to the gym' or 'you're fat,' very few have actually messaged me for the car itself."

The Jeep has been on the market for 19 weeks so far.

• Business Insider asked six tech workers which task they're saving the most time on with AI.
• Some workers said AI has turned tasks that once took hours into minutes.
• Others said the productivity gains haven't necessarily led to shorter workdays.

Ask a tech worker how AI has changed their jobs, and chances are they'll answer with a single number: hours saved.

In interviews with Business Insider, Big Tech software engineers, product managers, and data scientists described using AI to compress hours of work into minutes. They use it to draft documents, summarize months of meetings, review code, automate reports, and more.

Faster doesn't always mean easier, however. One Amazon data scientist said AI is adding hours to his workweek as he builds the automation systems that should eventually save him time. Another Amazon employee said any time saved is quickly redirected to the next project.

Here's how six tech workers said AI is saving them the most time. (Their responses have been edited for length and clarity.)

The time AI saves me gets reinvested into the next problem

Priyanka Devi Ramesh is a business intelligence engineer at Amazon. She's 30 and lives in Virginia.

Document writing is where AI has had the greatest impact. With the help of an AI tool called Pippin, it's become easy to translate my thoughts about the projects I'm working on into polished documents that can be technical or customer-facing. This saves a massive amount of time — I spend hardly 15 to 20 minutes max to write and finalize a document that would have previously taken me well over an hour.

On the technical side, I use Kiro and Amazon Quick. Kiro is great for brainstorming ideas and making logic updates in minutes. I'm building agents within Amazon Quick to automate common customer questions about dashboards and to surface insights from data.

AI hasn't reduced my work time. We're constantly looking for ways to clean up messy data and finding opportunities to automate wherever possible — so the time saved in one area gets reinvested into the next problem.

AI helps me make sense of months of meetings at Google

Prerit Pathak is a security engineer at Google. He's 27 and lives in New York City.

I use Gemini for a variety of purposes, but recently it has bolstered my note-taking.

I used to take shorthand notes during meetings to record interesting or important information. Now, I let Gemini take notes on my work calls, and the improvement has been incredible. A summarization task — such as understanding what happened over the previous six months — that once would have taken one to two hours now takes five to 10 minutes.

I'm working longer hours now, so AI can save me time later

Sarthak Gupta is a data scientist at Amazon. He's 29 and lives in Seattle.

AI has been most helpful with building end-to-end automation pipelines for recurring workflows.

It used to take 8 to 10 hours over a couple of days to create a monthly stakeholder report that involved pulling data, cleaning it, generating visualizations, and writing the summary. Now, an AI pipeline handles the data pulls, transformations, and dashboard refreshes. I spend maybe 45 minutes reviewing the output and adding context before sending it out.

However, my overall working hours right now are running longer than normal. The reason is that we're in the middle of an automation phase. Building the pipelines, integrating the AI tooling, validating outputs, and onboarding all of this into existing workflows is front-loaded work, and that upfront investment is real. The payoff comes later, when the same task that took a couple of days collapses into a button click.

So in the short term, AI is actually adding hours to my week, not subtracting them. I'd expect that to flip once the foundational pipelines are stable and the automation is doing the heavy lifting on its own.

AI helps me turn messy ideas into polished plans

Tanvi Pisal is a UX designer working as a contractor for Apple via Red Oak Technologies. She's 29 and lives in San Jose.

One of the biggest ways AI saves me time is in early-stage product thinking and documentation.

As a product designer, I used to spend hours drafting product requirement documents, brainstorming user stories, mapping edge cases, outlining use scenarios, and refining ideation before I even got to visual design.

Now, I can start with rough notes or a messy draft, and AI helps turn that into a much more structured document in minutes. What used to take me three to four hours can often be reduced to 30 minutes with feedback and refinements.

AI gets me to the starting line faster at Amazon

Udit Mehrotra is a head of product at Amazon. He's in his 30s and lives in Seattle.

Writing product documents is where I've seen the biggest change. Every major initiative at Amazon starts with a written document, and for years, the first hour or two of that process was building scaffolding: setting up the structure, filling in the sections you know by heart, and building something worth reacting to before you could get to the actual thinking.

Now I can use AI to input the customer problem and constraints and get a solid first draft in minutes. What surprises me is that it's often more comprehensive than what I'd have written on my own under time pressure.

Getting from 80% to 100% is still where the real work lives, and AI doesn't change that. The strategic judgment, the tradeoffs between what customers need and what's technically feasible, the decisions that require years of accumulated context about a specific customer problem — that thinking still takes the same depth and care it always did.

What has changed is that I arrive at the starting line faster, with a more complete structure to react to and push against. The quality of the final document is often better as a result, not because AI did the hard thinking, but because I spent more of my time on it.

What used to take a week can now take a day

Iren Azra Zou is a software engineer at the trucking logistics startup Double Nickel. She's in her 20s and lives in New Jersey.

I use AI, mostly Claude Code, for the majority of my coding. It's honestly hard to quantify the time savings; it feels like what used to take a week can now take a day.

We also rely heavily on AI to review and provide feedback on code, unless a change is particularly risky. That alone saves a huge amount of time. Instead of waiting days for human reviews, you get multiple rounds of feedback within hours. It also means I spend less time reviewing others' code, which probably saves me several hours each week.

There are tradeoffs — less human review can have downsides. But right now, the speed of iteration and innovation is incredibly valuable for us.

Do you have a story to share about how you're navigating a career crossroads? If so, please reach out to the reporter via email at jzinkula@businessinsider.com, or via Signal at jzinkula.29.

• Ukrainian President Volodymyr Zelenskyy wants to partner with Silicon Valley.
• Zelenskyy said the tech hub's AI skills and Ukraine's wartime drone experience could be "powerful."
• Ukraine has built a drone arsenal that's captivated the world as it fights Russia's invasion.

Ukraine has experience fighting and defending itself using drones. American tech companies have AI firepower. Ukrainian President Volodymyr Zelenskyy says the combination of the two could be world-changing.

"American technological companies have a lot of different interesting AI technologies that we don't have. And we have a lot of things that they don't have because of our experience on the battlefield," Zelenskyy said on CBS News' "Face the Nation" on Sunday. "I think this cooperation can be huge and the most powerful in the world."

Ukraine, out of necessity, has built an arsenal of drone tech and anti-drone tech on a shoestring budget, captivating the global defense industry as it's largely held the line — despite its underdog status — since Russia launched its full-scale invasion over four years ago.

Ukraine said three types of homegrown drones allowed it to strike in the vicinity of Moscow earlier this month, and that it had developed a fixed-wing mid-range attack drone that's helped it strike in areas Russia once deemed safe. It's learned valuable lessons in the process, like the need for drone units to always be on the move and for their command centers to be buried underground to protect them.

The AI craze in the United States, meanwhile, coupled with a Defense Department eager to develop new autonomous military technology, is fueling the growth of a Silicon Valley defense tech industry. Companies like Anduril, led by Palmer Luckey, who built the Oculus virtual reality headset that Facebook bought in 2014, have raised billions to develop new uncrewed weapons systems.

Ukraine has since become an important potential proving ground for some of that new hardware.

Through a state-backed "Test in Ukraine" program launched last year, hundreds of international companies have applied to test drones, counter-drone systems, AI, electronic warfare tools, naval drones, and ground robots in Ukraine.

Zelenskyy on Sunday said he wants to deepen this symbiotic relationship further, and soon. His message to Silicon Valley: Stop talking and start building.

"We need to negotiate already," Zelenskyy said Sunday. "Not to speak about it. Just to take steps and to do it as quick as possible."

• Apollo's chief economist said there's "zero evidence of AI-related job losses."
• A parade of tech leaders celebrated that take over the weekend.
• At least a dozen major companies, meanwhile, have cited AI in their decision to lay off workers this year.

Anyone worried that AI will replace them should take a deep breath, at least according to Apollo Global Management's chief economist.

In a blog post on Friday, Torsten Sløk said there is "zero evidence of job losses because of AI," citing the ADP National Employment Report. Instead, he said, companies are hiring candidates who have AI skills.

"Many firms are hiring AI implementation experts, and the data center buildout is putting upward pressure on salaries for AI experts and on prices of semiconductors, equipment, and energy," Sløk said. "The bottom line is that the AI spending boom is stoking both employment and inflation."

Sløk echoed that sentiment in an April blog post, writing that "cheaper inputs don't shrink industries. Instead, AI is going to increase both productivity and employment."

The latest ADP report found that private companies added almost 110,000 more people to their payrolls in April.

Anxiety that AI will eradicate the average job is everywhere, stoked in part by those behind the technology. While Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have recently changed their tune as they gear up for their respective IPOs, they have both warned for years that AI could upend entire job categories. Amodei famously said last year that AI could wipe out half of all entry-level white-collar jobs.

Sløk's analysis resonated with some figures in the AI industry, including Box CEO Aaron Levie, Dell CEO Michael Dell, and White House AI and Crypto Czar David Sacks, who all agreed with his view in X posts over the weekend. David Solomon, CEO of Goldman Sachs, also made a similar argument last week in a New York Times opinion piece.

An EY survey of 240 financial service CEOs, meanwhile, found that about 60% thought investing in AI would maintain or increase their staff head count in 2026.

These optimistic takes, however, seem to clash with recent reality. At least a dozen major companies have cited AI as a factor in staff layoffs this year. In February, Block CEO Jack Dorsey said the company was slashing its workforce from over 10,000 to under 6,000.

"We're already seeing that the intelligence tools we're creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company," Dorsey said in a memo shared to X. "i had two options: cut gradually over months or years as this shift plays out, or be honest about where we are and act on it now."

Cisco, Atlassian, Cloudflare, Coinbase, IBM, and Snap are also among the companies that have cited AI as a reason for layoffs.

Nvidia CEO Jensen Huang, one of the pillars of the AI industry, has criticized companies that blame AI for layoffs. "I think the narrative that connects AI to job loss for many of the CEOs that are doing it is just too lazy," Huang told a media outlet in Singapore last week.

Altman has called the practice of blaming AI to reduce staff "AI washing."

In his blog post on Friday, Sløk said that, in his view, the current employment climate is an example of the "Jevons paradox," an economic theory that says as new technology increases the efficiency of a resource, the more that resource is consumed.

In this case, that resource would be human workers.

"It is Jevons paradox playing out in real time: cheaper technology is creating more demand and more jobs," Sløk wrote.

• Environmental activist Erin Brockovich has adopted a new cause: the impact of data centers.
• She said residents are especially angry about NDAs between developers and local leaders.
• That lack of transparency, she said, is fueling anger among residents who feel ignored.

Big Tech is expanding into communities across the country — and they aren't all that happy about it.

Many residents in cities and towns where tech companies are looking to build large data centers to power their AI products are mobilizing against them, concerned about a possible drain on water supplies, a surge in electricity costs, and a decline in their overall quality of life.

Now, legendary environmental activist Erin Brockovich, famously played by Julia Roberts in the 2000 film about her work, has joined the fight.

Brockovich said on a recent episode of "The Jim Acosta Show" that communities are angry because they feel shut out of the decisions being made in their own backyards — and that the projects are being "shoved down their throat in secrecy."

Brockovich said that residents learn about projects in the proposal stage, only to find that local officials are limited in what they can say because of nondisclosure agreements. In other cases, she said, projects are presented as warehouses rather than data centers.

"There's a lot of secrecy and NDAs at a very proposal stage," Brockovich said.

That lack of transparency, she said, is fueling anger among residents who believe their concerns are being ignored.

High-profile data center projects have faced backlash in recent months. A massive data center project in Utah backed by "Shark Tank" investor Kevin O'Leary has sparked statewide opposition, for example, leading Utah Gov. Spencer Cox to unveil a new "framework" for data center development on Friday that addresses many of the community's concerns.

"Utahns deserve confidence that water resources, air quality, utility rates, wildlife, and quality of life will be protected. This framework helps ensure that data center development aligns with Utah's long-term interests and reflects Utah values," Cox wrote in an X post.

Microsoft, which once relied on NDAs in the early stages of data center development, said earlier this year that it would stop requesting them after local opposition.

"We've made the decision that being transparent with the communities where we operate or seek to operate is paramount," the company said. "This shift is about strengthening public trust, enabling better dialogue, and ensuring that our growth is matched by meaningful engagement."

Microsoft has adopted its own framework for building data centers called the "Community-First AI Infrastructure Plan." It promises to pay for its own electricity, minimize water usage, and create local jobs, among other things.

Brockovich, who has spent decades working with communities on environmental fights, said residents are not opposed to hearing difficult information. What they object to, she said, is being excluded from the process.

"I've worked in communities for 30 years," she said. "They handle the truth."

O que uma estrela Michelin pode mudar na vida de um chef de cozinha já consagrado? O efeito mais imediato é a visibilidade. E isso vale para quem ganhou a primeira estrela e para quem passou de duas para três, como os restaurantes Evvai, de Luiz Filipe Souza, e Tuju, de Ivan Ralston, em São […]

O post Entre o brilho e a pressão: três chefs consagrados revelam o que muda com novas estrelas Michelin apareceu primeiro em NeoFeed.

• Hana Elster, a college senior, used Claude during winter break to create a website in five days.
• She spent under $2,000 to get an online vintage marketplace up and running.
• Elster said she hopes that it'll be a successful side hustle as she moves into the corporate world.

This as-told-to essay is based on a conversation with Hana Elster, a 22-year-old senior at Boston University who founded VYA, an online vintage marketplace. It has been edited for length and clarity.

I'm a senior at college studying business law at Boston University, graduating in May.

I've always loved the idea of being your own boss and building something for yourself.

I'm also surrounded by a big founder community in Boston and New York. Seeing all these other people my age or a little bit older building things for themselves inspired me to do something for myself.

Over winter break, I used Claude to build an app that now has hundreds of users.

My love for vintage gave me a business idea

I went thrifting for the first time when I was around 14. You'd go to the regular mall and find a shirt for $30, but I could find 10 shirts for $30 at my nearest Goodwill, and I'd be wearing something nobody else had.

So it piqued my interest in secondhand shopping, getting good value, and developing a unique style. Now, I would say 80% of my closet is secondhand.

Last year, I was talking with one of my friends about our favorite stores in our hometowns, and she told me about an amazing vintage store in Chicago. I'm originally from Washington, DC, and I wouldn't have known about this store unless we'd had this conversation.

I realized there is space for a centralized vintage platform.

I reached out to many shops to learn about their pain points and where they want to grow. The sellers mainly said they want more eyes on their products.

Brick-and-mortar vintage stores told me they had websites, but most of their sales come from foot traffic. Shop owners said they put a lot of work into their website, yet they only got one sale a month on that platform.

Stores that are fully online said they had to post three times a day on Instagram and TikTok because marketing was the only way to get their name out. A lot of these sellers also had full-time jobs, and vintage sales were their side hustle, so they didn't have time to do much work.

Using Claude to code a website in five days

To build the app, I started vibe coding. I have some coding experience, but Claude Code has let me move at a pace I would never have been able to.

I connected with a friend who had built a fashion app years ago, and he prompted me to learn how to vibe code. In January, I started with Cursor and eventually began coding with Claude, and the project accelerated really fast.

I started on January 9 and built a mock website for VYA by January 13, over winter break. It got me so excited, thinking about how I was turning lines of code into something visual, with buttons and functions and everything.

Five days later, I had a website. Friends in tech also helped me look over the code and make sure everything flowed correctly.

Monetization

I've spent under $2,000 on app development so far, which covers all technical and operational costs. It's self-funded by my savings, and I've also received some grants from Boston University.

More than three months on, we have 38 stores fully onboarded and continuing to grow, and roughly 900 approved users, of whom 50% are daily active users. My goal is that when you're checking out vintage websites like The RealReal, you also open VYA.

To monetize it, I'm charging a 7% commission per item sold, with an average price of about $350. I'm also trying out a model called "source for requests," in which we charge customers a flat fee for finding a particular product, like a rare 90s Chanel bag.

Usually, one of the 38 stores can fulfill the order, and if we can't, the fee is refunded.

AI has both dropped and raised barriers

I've always wanted to start my own business, but this was the first time I was hit with inspiration, and I could actually execute it. I don't know if I would've done it if I'd known I'd have to hire engineers and other staff, because I would've had to raise money.

So AI has definitely dropped the barrier to entry.

The biggest barrier used to be engineering, but now it's getting people to hear about your brand and then convert and buy, which is why my first hire is going to be a CMO.

I'm thinking of how to raise money so that I can add to my head count. I'm focusing on growing and getting more people to help me out so that I can grow exponentially faster.

After graduation, I'm supposed to work in consulting — I've got a return offer, and the role doesn't start until September.

I'm planning to accept it and see how this business grows on the side. If I can build it up enough, grow my head count, and automate it, I would love to do it alongside my corporate job.

Being a young founder has changed me. The other founders I've seen are super bold, confident, and courageous, and I feel like I've developed that side of myself, too.

• Mark Zuckerberg's email struck an empathetic tone. He also said he didn't expect more companywide layoffs in 2026.
• Layoff anxiety can hurt worker productivity and morale, thereby carrying a real business cost.
• Workplace observers say his focus on stability suggests he recognizes the impact of prolonged uncertainty.

Mark Zuckerberg is signaling that Meta employees can stop looking over their shoulders.

After long emphasizing cost-cutting, management flattening, and "Year of Efficiency" rhetoric, the Meta chief struck an empathetic tone in his post-layoff email to employees on Wednesday — emphasizing stability, conceding communication failures, and promising to "do right by people along the way."

In his internal email to staffers, he thanked the roughly 8,000 workers who were being let go and emphasized his desire to provide "as much stability as possible" to those who remained.

It was a reminder that layoff anxiety carries a real business cost.

To that point, Zuckerberg said that he doesn't expect further companywide layoffs in 2026.

While that doesn't rule out smaller-scale cuts, the message followed weeks of grueling uncertainty for staffers waiting to learn whether they still had jobs.

Zuckerberg's email — a shift away from the more hard-charging tone he adopted post-pandemic — suggested he recognizes that prolonged uncertainty can weigh on employees and, ultimately, the company itself, workplace observers told Business Insider.

"You do need to try to create some psychological safety for people who are there, because layoffs are extremely distracting," said Amii Barnard-Bahn, a C-suite coach and consultant.

'We won't always get this balance right'

Wednesday's cuts were the latest challenge for a workforce that has spent years navigating repeated rounds of layoffs, heightened performance scrutiny, and persistent questions about whether AI would take their jobs.

It's a theme that has played out across tech, as companies increasingly tie cuts to AI and leaders warn about a white-collar bloodbath.

In 2025, the CEO told staffers in an all-hands meeting to "buckle up" for an "intense" year ahead. Some of Meta's layoffs have come with an added sting: Last year, the company also said it was cutting some 4,000 workers who had failed to meet expectations.

By the time the latest round arrived, the accumulation of uncertainty had drained some employees and left them wishing they were let go.

Meta didn't respond to a request for comment from Business Insider.

Zuckerberg's Wednesday message hit on the toll that uncertainty around staffing levels can take: "We won't always get this balance right, but I care deeply about this so we'll keep adjusting and work hard to do right by people along the way," he wrote.

It's not clear how effective Zuckerberg's softer tone might be, though he had little choice but to try to reassure those left standing, said Pav Stojkovic, an HR consultant and former chief people officer at several companies, including The Athletic.

Zuckerberg's approach is a departure from one he'd used previously. In 2022, for example, Zuckerberg told Meta staff he was upping performance goals to get rid of employees who "shouldn't be here."

By "turning up the heat a little bit," Zuckerberg said at the time that he hoped some workers would "decide that this place isn't for you, and that self-selection is OK with me."

Last year, Meta directed managers to place a higher proportion of employees in its bottom review rankings. Zuckerberg has a long-standing history of ratcheting up the pressure at Meta, reinforcing a blunt, survival-of-the-fittest culture at the social media giant.

The billionaire CEO is far from alone in embracing a sink-or-swim philosophy as AI reshapes the workplace.

A focus on execution

Zuckerberg's note comes at a transitional time for the industry. Excitement over the possibility of AI has mixed with fears over efficiency-driven job cuts and the encroachment of automation on workers' livelihoods.

As Meta reshuffles roughly 7,000 employees to focus on new AI initiatives, Zuckerberg needs a workforce concentrated on execution amid the AI arms race.

"Success isn't a given. AI is the most consequential technology of our lifetimes. The companies that lead the way will define the next generation," he wrote.

Barnard-Bahn said it's likely that productivity at the company took a big hit in the last month, as workers worried about whether they or their colleagues would be cut or reorganized.

By providing workers with a higher degree of job security for the next six-plus months, Zuckerberg might be offering employees something that Big Tech competitors have not.

"Meta has the talent, the infrastructure, the apps and distribution, and the business model," Zuckerberg wrote. "We have a lot of work ahead, but what's on the other side is going to be extraordinary."

• Steve Wozniak gave a commencement speech at Grand Valley State University earlier this month.
• He was cheered after telling students they had AI, or "actual intelligence."
• Students have booed some other execs who championed AI during their graduation speeches.

Steve Wozniak did what other college graduation commencement speakers couldn't this year: earn applause when talking about AI.

The Apple cofounder took the stage during Grand Valley State University's graduation ceremony earlier this month. During his speech, Wozniak offered reassurance to new graduates who are entering the workforce at the height of the AI revolution.

"You have AI — actual intelligence," Wozniak said.

The remark garnered laughs and applause from the audience.

"It would take too long to go deeply into what I think about AI, but we've been trying to create a brain," Wozniak said. "Is there a way we can duplicate a routine a trillion times and have it work like a brain? AI is one of those attempts."

While Wozniak delivered his speech without interruption, the same can't be said of some other AI-forward commencement speakers. In the weeks that followed, former Google CEO Eric Schmidt and real estate executive Gloria Caulfield were both booed for their comments about AI at two separate graduate ceremonies.

AI is looming over new graduates as they enter the job market. The tech is changing the landscape, from the skills candidates need to how companies assess them. Its ability to automate many tasks has led some companies to conduct AI-related layoffs.

During his commencement address, Wozniak reflected on working at Apple and offered students some advice as they begin their careers.

"You should always try to think different," he said. "Don't follow the same steps as a million other people. Think, is there something I can do a little different?"

• Business Insider is compiling its third annual list of VCs funding social startups.
• We want to hear from you about who the standout VCs are in 2026 — and where they're placing bets.
• Please submit your ideas through this form (or below) by June 4.

With AI taking over feeds and the biggest social media platforms feeling less social than ever, there's an opportunity for a breakthrough startup.

Business Insider wants to know: Who is backing those companies?

Some venture capitalists are on the hunt for the "next big thing" in consumer technology — even as VC interest in the broader consumer category is in flux.

There are new social networking startups vying to be the next generation's Facebook, Tumblr, or TikTok. The space is also brimming with new companies trying to disrupt dating and day-to-day life with AI assistants that help you navigate your in-real-life relationships and plans.

We're publishing our third annual list of investors funding innovative startups building technology that's advancing the internet, and meeting the needs of people who want to feel more connected (whether that's online or IRL).

In 2025, the investors Business Insider highlighted were placing bets in categories like IRL social, vibe coding, and platforms for niche interests or communities.

This year, we want to know which leading investors are backing social and consumer-facing internet startups, and where they are placing their bets.

Please submit your ideas through this form (or below) by June 4:

• Jamie Dimon said he'll probably hire "more AI people" and "less bankers" in the future.
• The JPMorgan CEO said the bank is in the early innings of using the fast-moving technology.
• He also said a fellow bank CEO's controversial statements about AI job losses were "inartful."

The future of the nation's biggest bank might mean fewer bankers.

JPMorgan CEO Jamie Dimon said that AI will impact "every job," and change the balance of his more than 300,000-person workforce in an interview with Bloomberg that aired on Wednesday.

"I think it'll reduce some of our jobs down the road," he said. "I think we'll be hiring more AI people and probably less bankers in certain categories."

He agreed that AI efficiencies will lead to downsizing when asked, but said that similar patterns have been happening throughout his life. Dimon said that JPMorgan sees about 10% attrition each year, equivalent to around 30,000 people, and the bank is prepared to reskill them, give them new jobs, and potentially offer early retirement. During the company's investor day in February, he said that they already have "huge redeployment plans."

"We can take people who are displaced — and we have displaced people from AI — and we offered them other jobs," he said at the event.

During Wednesday's interview, Dimon said that JPMorgan is using AI for everything from risk to marketing to coding, and that's just "the tip of the iceberg" as the technology rapidly accelerates. The bank has a $20 billion technology budget and already keeps careful tabs on how its engineers use AI.

Aside from numbers, the very nature of banking is already changing, as startups like Rogo and Hebbia automate some of the grunt work that previously defined junior roles. Anthropic recently rolled out a suite of AI agents for the financial sector, including ones for building notoriously tedious pitch decks and models.

Dimon also addressed Standard Chartered CEO Bill Winters' controversial comments about AI-related job losses. Winters described his planned reduction in support staff as "replacing in some cases lower-value human capital with the financial capital and the investment we're putting in." In the wake of an online backlash, Winters clarified his language in an internal memo on Wednesday, writing that "where roles do fall away, it reflects changes in the work, not the value of our people."

Dimon called Winters, who spent 26 years at JPMorgan, a "friend," and said that we've all phrased things poorly.

"It was an inartful way to say something," Dimon said, before adding that AI will impact everyone, not just less-skilled employees.

When it comes to employee well-being beyond the office, Dimon said he urged the importance of keeping New York City competitive in a recent meeting with Mayor Zohran Mamdani.

"We went through that every city has to compete, and they have to compete at every level," Dimon said of his conversation with the mayor, who has recently faced criticism from much of the business community for singling out Citadel CEO Ken Griffin in a video.

Dimon said that the city's high taxes are "already" causing talent to leave, pointing to his own growing workforce in Texas.

• Google unveiled a new AI-powered search this week. It could ruin the internet.
• The changes mean you'll get a traditional list of links much less often.
• Instead, you'll get AI-powered answers, personalized just to you. That's not the internet we know.

Google is about to ruin the internet.

And I love the internet. I love websites. I love sending links to my friends. I spend nearly my entire workday looking at various Chrome tabs. I enjoy looking at websites I've never looked at before.

All of that is going to change with Google's new search updates, which lean into AI-generated answers. Its plans gave me an awful sinking feeling.

The big change will be to integrate AI even more into search — instead of typing in a few keywords and getting a list of links, Google will spit out more AI-powered answers to questions and personalized requests (I'll explain more in a bit.)

When a Big Tech company gives a demonstration of a new product, you've always got to look at it with some skepticism, so let's take this with at least a grain of salt: But Google, this week, said its new features would let an AI agent send you updates whenever your favorite athletes launch a new sneaker, for instance. (I am not sure how it knows your favorite athletes, but, at the same time, of course, Google knows that about you.)

A personalized internet isn't the internet

I really struggled to get into the mindset of someone who wants this. I'm not a sports fan or a sneakerhead, so new athlete-endorsed shoes don't thrill me, but I do enjoy shopping, so I can see the appeal of buying cool new sneakers. Do I want an alert when an athlete or brand announces this? Do I expect this to be something I wouldn't see in a TV ad, read in a style magazine, or see on the athlete's Instagram? Like, I guess I can see some convenience here, but there's already a robust ecosystem of ways to get this information that I find generally pleasurable to engage with.

Another example: Someone asking Google for advice on where to go for a hike with restaurants and parking nearby. That's a nice, wholesome project for a demonstration, but I can't imagine using Google in this way. I use Google in that time-honored way we've all been taught: typo-laden, two-word utterances — not elegantly worded requests.

I usually already know what I want to find; I just need help getting to the website that gives me the thing I want.

Google, of course, is bullish on its new product — and especially the personalization of it all. "We believe the best version of search is one created just for you," said Robby Stein, vice president of product for Google Search, during the I/O presentation this week.

I guess? Sure?

Stein gives an example of a college student asking about black holes. The student types a question in the search bar, and Google gives an AI-generated answer and creates a customized animation that shows how black holes work. (Eventually, it will give you a few links to other information about black holes.)

OK, cool. But that's not really what "search" is, right? That's just an AI chatbot that answers highly specific questions.

'Google Zero' might be coming

All this is obviously potentially very bad news for websites that have depended at least in part on Google search traffic, including the one you're now reading. There's been an expectation of this coming eventually — a doomsday event referred to glumly as "Google Zero" when Google traffic, which has been dropping across the news industry for the last few years, eventually hits zero. AI Overviews and people using other AI tools like ChatGPT to find information have already cut deeply into publishers' search traffic.

I am aware that my aversion to this new search experience may seem like self-preservation, since this is ostensibly not good for the journalism industry (in the short term, at least). But I assure you that my complaints are more personal and short-sighted. I am annoyed that this will change how I search and how I experience the internet.

I think of "the internet" as a place you go to; it shouldn't come to you.

I know it's outdated, but imagine the information superhighway — a physical space of roads, silos, and dungeons you can wander around. Google is the doorway to it. It hands you the map so you can give yourself a self-guided tour. But you're on your own in there.

By doing this for, say, 12 hours a day, seven days a week, for, say, two decades, you become quite adept at knowing where to look and how to navigate. I still think of the internet this way.

The new Google seems like a gated community

But a version of the new AI search is one where you never have to venture out onto the internet; it's brought to you in a sanitized form by an intermediary. It's like living in a gated community with a strict HOA vs. a walkable city with public transit.

I know that harping on the virtues of "websites" is something only old people do; a leftover of the techno-optimism of people who spent time on old-timey messageboards or blogs. And, well, sure, I'm an old crank, but I've also got the youth on my side on this one. As my colleague Dan DeFrancesco writes, Gen Z has an aversion to AI — and this may affect how they adopt these new Google search products. I say, if Gen X fads like hacky sacks are back, then so should caring about the open web.

I don't want Google search to trap me into its AI HOA! I want my internet experience to have a tacky-colored house, an unkempt lawn, and a year-round 12-foot Home Depot skeleton. That's what the internet should feel like to me: something to explore and venture out into, even if that means a tiny bit of imperfect friction.

A startup brasileira Sinatra AI, que usa inteligência artificial para reduzir perdas no e-commerce, movimentou R$ 10 milhões em uma rodada seed. O aporte foi liderado pela BluStone e teve participação de Caravela Capital, GR8 Ventures e Plug and Play. Além da injeção de capital, parte dos recursos da rodada foi destinada à compra da […]

O post O que a BluStone viu na Sinatra AI, startup que caça erros no e‑commerce apareceu primeiro em NeoFeed.

• Meta previously announced it will cut 10% of its staff next month.
• Meta's HR chief told staff in a meeting that she can't promise further layoffs won't happen.
• She added that the business is strong and acknowledged that morale has been affected at Meta.

Meta plans to lay off around 10% of its staff next month, and it told staff it's not ruling out deeper cuts.

That's what Janelle Gale, Meta's chief people officer, told employees in an internal meeting on Thursday, according to three sources on the call.

"Will there be more layoffs? The question always comes up. I'd love to say that there are no more layoffs, but I can't say something we can't deliver," Gale said during the meeting. "While the business is strong, priorities change, competition is fierce, and we will continue to manage our costs responsibly."

She said this means that Meta will "continue to evolve teams as needed" and "try to redeploy talent." She pointed to how Meta is investing in its Applied AI organization.

Gale added that some organizations would be more affected by layoffs than others, though she did not specify which.

Meta leaders also said during the meeting that AI token usage would not be considered as a factor for the layoffs.

Meta CEO Mark Zuckerberg also addressed the layoffs at the meeting, saying that AI automation is not the driving factor behind them. He said that AI has made small teams far more efficient.

During the call, Zuckerberg also addressed Meta's plan to monitor employees' keystrokes and mouse movements to improve its AI models. He said humans are not actually watching what the staff are doing and that this data is abstracted and used to improve AI.

Meta AI Chief Alexandr Wang also appeared at the meeting, sporting a camouflage-pattern T-shirt featuring multiple deer, according to a photo seen by Business Insider. During the Q&A, he praised Meta's latest AI prowess, notably the recent release of its Spark model.

Meta declined to comment for this article.

Reuters reported in March that Meta plans to cut about 20% of its total staff this year.

Given the looming layoffs, Gale said at the meeting that they hit morale at Meta, and the company tries to make tough situations like that "the best version possible." She added that Meta has tripled COBRA healthcare coverage to 18 months.

Meta CFO Susan Li previously said during its first quarter earnings call on Wednesday that she "doesn't really know" the ideal size of the company's head count, which runs at above 77,000. Meta announced that its infrastructure spend, largely for AI, is doubling this year, to a range of $125 billion to $145 billion.

Have a tip? Contact Charles Rollet via email at crollet@businessinsider.com or on Signal and WhatsApp at 628-282-2811. Contact Hugh Langley via email at hlangley@businessinsider.com or Signal at 628-228-1836. Use a personal email address and a non-work device; here's our guide to sharing information securely.

• A personal injury lawyer apologized for filing court documents with fabricated quotations.
• The lawyer told the judge that he had begun using software from a venture-backed startup called Eve.
• The episode highlights a growing risk for the startups selling artificial intelligence to lawyers.

Lawyers keep getting burned by artificial intelligence that invents cases and makes up quotes. Now, some attorneys are naming the software they used.

Last month, a Louisiana personal injury lawyer apologized after submitting briefs that cited a real court decision but quoted passages that didn't exist. The mistakes appeared in two filings in the 19th Judicial District Court in Baton Rouge and were flagged by opposing counsel.

"I'm trying to understand how I made this mistake," Ross LeBlanc, a partner at Dudley DeBosier, wrote in a private letter to Judge William Jorden on March 27. Earlier this year, he said, he began using an artificial intelligence program called Eve to draft pleadings. At first, he checked the citations often. "They were always correct when I checked them," he wrote.

That consistency gave him confidence, and eventually, he stopped checking, he said.

"I never thought this could happen to me," LeBlanc wrote, adding that he could not be sure whether the mistake involved Eve's software or if he copied and pasted something too hastily.

Jay Madheswaranm, Eve's chief executive, told Business Insider on Thursday that after a close audit of the case with Dudley DeBosier, the company confirmed Eve "did not hallucinate any case citations in this matter," including any fabricated quotations.

Courts have slapped sanctions on attorneys for filing briefs with errors created by artificial intelligence — often called "hallucinations." Last week, Sullivan & Cromwell, one of the country's oldest and most elite law firms, apologized to a federal judge over a similar slip-up.

What's new here is the blame game. When an attorney names the tools involved, the companies behind the software are put in the spotlight and could face reputational repercussions.

Legal software companies like Harvey, Legora, and Eve have raised billions of dollars on the promise that they can make lawyers faster — and offer firms a level of reliability that general-purpose tools can't match. If their software starts to embarrass customers in court, that trust erodes.

Damien Charlotin, a French researcher who tracks hallucinations in court filings, estimates that fewer than 10% of cases identify the software used. Many lawyers, he suspects, keep that part private because they're relying on free chatbots like ChatGPT or other off-the-shelf tools that may not be authorized for client work.

Last year, a Latham & Watkins lawyer defending Anthropic in a copyright lawsuit made headlines after citing an article that does not exist. The lawyer said the mistake stemmed from using Anthropic's own chatbot, Claude, which fabricated an article title and authors.

Eve builds software for plaintiff-side lawyers using large language models, helping them draft documents, map out medical histories, and send and respond to discovery requests. The company was valued at $1 billion after it raised a $103 million funding round about a year ago. Madheswaranm said Eve now processes more than 200,000 documents and other results a month — up around 100-fold from a year ago.

LeBlanc told the judge that he had been wary of the technology generally because of the "horror stories" about hallucinated case law. He said he was persuaded after Eve pitched the tool to his firm and assured attorneys it had safeguards to reduce errors. He believed the risk was limited as long as he conducted his own legal research and directed the software to rely only on approved sources.

Then, opposing counsel in the personal injury case pointed out his mistakes.

LeBlanc's apology surfaced this month in a separate case involving a trip-and-fall at a Lowe's store. The opposing counsel found hallucinations in a brief filed by Dudley DeBosier and included LeBlanc's letter in a request urging the court to expand its inquiry into possible sanctions.

Dudley DeBosier has filed a motion to strike opposing counsel's request because it says the cases are unrelated. The firm also indicated that a lawyer used Claude to help draft the brief in the Lowe's case.

It's a view widely shared across software companies and law firms that artificial intelligence can assist in research and drafting, but responsibility for the final product remains with the human who signs the filing.

Madheswaran said Eve makes that explicit in its contracts and onboarding with new customers. The software also includes features designed to catch errors before they reach a courtroom, though they don't always work. Some errors are harder to spot than others, he said. Confirming a case exists is easier than verifying a quote is exact.

As the legal profession races to adopt artificial intelligence, mistakes are more likely to be caught. Courts are getting wiser to the technology, and opposing counsel are adjusting their tactics. Instead of only attacking legal arguments, lawyers are scanning filings for errors that could undermine the other side's credibility.

Chad Dudley, a founding partner of Dudley DeBosier, a firm with about 40 attorneys, said it trains its lawyers to carefully review generated results and requires them to agree to use the technology responsibly.

For his part, LeBlanc said he hopes other lawyers learn from his mistake. He told Business Insider on Thursday that Eve helped him move faster under time pressure, but after the errors surfaced, he felt "sick to my stomach" and couldn't sleep.

"I'm responsible for checking everything, no matter what technology comes along," he said.

He doesn't blame Eve for the blunder. Still, he's setting the tool down for now.

"I feel like, given what happened," he said, "it's fair to have a cooling off period, you know, touch grass."

Have a tip? Contact this reporter via email at mrussell@businessinsider.com or Signal at @MeliaRussell.01. Use a personal email address and a non-work device; here's our guide to sharing information securely.

• Silvio Pfeufer uses Saltz to save time on food procurement at his Michelin-starred restaurant, Matthias.
• Saltz uses agentic AI to modernize a typically complicated process between food buyers and sellers.
• For independent eateries, the streamlined process can reduce costs and increase access to high-quality ingredients.

When Silvio Pfeufer first got into cheffing, he was surprised by the amount of administrative work involved.

"It's not only the evening, sending nice plates to guests. There's a lot of stuff to do to make that happen," Pfeufer told Business Insider. He took long phone calls from food suppliers and producers that kept him away from the kitchen.

Sourcing and buying food is a process riddled with disorganization, since farmers, wholesalers, logistics providers, and restaurants communicate in different ways — by phone, email, text message, and PDF blast — starting as early as 4 a.m. The cost and availability of products are constantly in flux, so sellers issue frequent updates: The price of wild-caught fish could change three times a day, for example.

When Pfeufer opened his own restaurant in 2024, he wanted to streamline his food procurement process. At Matthias, which pays homage to his late grandfather, Pfeufer uses an AI-assisted platform for food business owners called Saltz to speed up food procurement and reduce phone calls.

The time savings are critical, said Pfeufer, a co-owner and head chef at the Berlin-based eatery. Using AI agents, Saltz connects restaurants directly to suppliers via a marketplace that brings together disparate catalogs, transactions, and logistics so restaurants can compare and buy fresh, high-quality, and specialty products more quickly than with traditional food procurement processes.

The platform's standardized, real-time food data can be a game-changer for independently owned and operated restaurants like Pfeufer's, given their limited purchasing power. Saltz said that thousands of buyers and hundreds of suppliers use its technology, though it didn't share exact numbers. It said around 80% to 90% of its buyers are independent restaurants.

Connecting with high-quality and specialty suppliers

Pfeufer said he chose Saltz because he liked how the platform modernized old processes. In the two years he's used it, it's allowed him to discover new suppliers, order outside business hours, and gain better oversight on pricing, he said.

Founded in 2022, the startup uses AI agents to ingest PDFs, emails, and text messages that food sellers send to Saltz, standardize and enrich all the data, and consolidate it into a single platform. Its AI agents automatically update each seller's listings on the platform, eliminating the need for manual updates.

Buyers, meanwhile, see once-disparate product options, information, and up-to-date prices in one place. They can also order food at a time that's convenient for them and track their deliveries. "On a Sunday, at night, or in the morning, I can do it by myself, and don't have to have all these calls," Pfeufer said.

At Matthias, which was awarded its first Michelin star in 2025 after 10 months of service, meals must meet a high standard every day, and products have to be of the freshest quality. Pfeufer orders vegetables, milk, and other items from Saltz every week, plus fresh fish twice a week. Before using Saltz, the quality of his products wasn't necessarily worse, he said, but access to new or specialist suppliers was more limited and supply chains were longer.

"The fish is now often sourced directly from the trader or farmer, without the need for intermediate storage. This allows us to avoid additional storage times that could negatively affect freshness," Pfeufer said, adding that more oversight into the supply chain — and it being shorter — means his food is more consistently high quality.

"We've definitely been able to connect with better suppliers," he told Business Insider. The chef added that he still works directly with certain local farmers, as he is often on the hunt for rare items that aren't on the platform.

For Pfeufer, using Saltz also allows him to see more costs upfront, so he can better plan how much to charge for new dishes. "It makes all the calculations much easier, which is very important for us," he said.

Agentic AI acts as a foundational tool

Saltz was founded by brothers Andrius and Thomas Šlimas, who previously built the Shopify-acquired dropshipping platform Oberlo. After spending four years inside Shopify's supply chain machine, the brothers teamed up with industry veteran Reinis Štrodahs. Their goal: modernize the $9.8 trillion food procurement industry.

"It's impossible to make sense or structure that chaos of information which lives in different places and has no common structure," Andrius said.

The Šlimas brothers said that previous unsuccessful attempts to modernize food procurement took two approaches: either trying to force suppliers and buyers onto a single platform, which required them to change how they work, or taking on the time-consuming task of manually inputting every PDF, email, and text message.

With Saltz, AI agents upload and update product listings for sellers, so individual stakeholder workflows don't have to change. Making agentic AI foundational to the process, Tomas said, gives their platform an edge.

"That gives us a speed advantage, and in this market, speed compounds into market share."

• Pinterest's VP of product management said mixing open-source and closed-source AI models helped cut costs.
• The social platform's recent AI-enabled features combine OpenAI, Alibaba, and its own AI models.
• To support this approach, Pinterest invested in cloud infrastructure and an updated hiring strategy.

In its pursuit of creating AI-supported tools for its users and advertisers, Pinterest leaders said it is actually shrinking part of its AI budget.

As companies continue to spend on AI, they face pressure from investors to show returns on investment from this fast-evolving technology. One way to do that: Stretch investments by shrinking the budget needed for the systems that power AI features.

Pinterest, the visual-first social media platform where users can save images to curate their ideas, is taking what it calls a "model-agnostic" approach to generative AI, said Vicky Gkiza, the company's vice president of product management. The strategy started in 2023 and involves combining Pinterest's proprietary AI models, developed by the company's software engineers, its closed-source models from Anthropic and OpenAI, and open-source models from Alibaba, said Gkiza.

While closed-source AI models can quickly process large amounts of data, require little maintenance, and integrate instantly into a company's existing systems due to their pre-built nature, they're typically more expensive. Open-source models, meanwhile, are typically free to download, use publicly available data, and can be modified more deeply than closed-source offerings. However, they also require the expertise of software engineers who can build, maintain, and debug these highly customizable models.

Lan Guan, the chief AI and data officer at Accenture, said companies are increasingly seeing value in a multimodal AI approach that balances performance and the cost of tokens, or the units of text AI models process. "This token cost is going to slow you down if you don't start managing them proactively," Guan told Business Insider. "Open-source will be a really good option."

Gkiza walked Business Insider through how Pinterest bolstered its blended approach to generative AI through beta testing, an updated hiring strategy focused on software development expertise, and infrastructure investments.

The tech

Pinterest uses OpenAI's closed-source large language models to support some of its product features and relies on Anthropic's Claude, another closed-source AI, for internal use cases like coding, a company spokesperson told Business Insider. Alibaba's Qwen, an open-source LLM, is used for visual and content understanding, data labeling, and assistant tasks.

With this approach, the company rolled out several new AI features in 2025, each blending open-source and closed-source generative AI, said Gkiza.

Auto-collages, a feature where advertisers can convert their product catalogs into pins that populate on shopping feeds for categories like skincare, home decor, and fashion. Pinterest began testing auto-collages at the beginning of 2024, when software engineers fine-tuned a mix of internally developed, open-source, and third-party AI models. By June 2025, auto-collages was ready for a pilot, which included a small group of retailer-advertisers like Macy's.

Then, Pinterest's voice-enabled AI feature, which uses both open-source and third-party AI to generate responses to users' queries, underwent beta testing in October 2025. Early results showed that users tended to pose more shopping-specific questions when they could ask aloud rather than typing their searches, said a Pinterest spokesperson.

"Search has been evolving so fast. It was imperative for us to use AI to improve," Gkiza told Business Insider.

The talent

To advance its new AI-enabled features and execute its blended-model approach, Pinterest also hired employees with AI and machine learning expertise to steward the customization of its large language models, said Gkiza. Former Google engineer Matthias Zenger joined the company as its vice president of engineering in April 2025. Three months later, in August, Pinterest announced it had hired software engineer Mirjam Wattenhofer to focus on e-commerce and user experiences, and that it would open an Engineering Excellence Hub in Zurich.

Both Zenger and Wattenhofer work at the Zurich office, where a team focuses on improving user experiences with AI and machine-learning technologies. In February, Pinterest CEO Bill Ready said the company would hire additional research and development workers to support the company's AI efforts.

"We are investing more in hiring the right talent — evolving the team, whether it is engineering or product management — to be much more familiar with AI," Gkiza said.

The outcome

Gkiza said the company's blended approach to AI-driven experiences costs an estimated 90% less than when Pinterest relied solely on its proprietary models, a milestone the company touted during its February 2026 fourth-quarter earnings presentation.

With its growing use of open-source AI models, the company plans to invest more in cloud-computing infrastructure, like graphics processing units, which are needed to power the cost-saving technology, Gkiza said.

Pinterest said it will continue to experiment with various AI models, prioritizing its own models for personalization, open-source models for cost-effectiveness and multimodal machine learning, and closed-source models when they still offer the best performance.

• OpenAI CEO Sam Altman sends a lot of messages.
• Altman said that he messages "a few hundred" OpenAI employees a day.
• Overall, he said that he's "definitely not a hands-on manager."

OpenAI CEO Sam Altman has a very direct way to get his message across.

In a conversation with Stripe CEO Patrick Collison, Altman said that he messages "a few hundred" OpenAI employees a day between texts, Slack, and other media.

"Very quick, like one, two messages, whatever, not done by an agent," Altman said during an appearance at Stripe Sessions. "I actually do it. And the context I get from that sometimes is very helpful in these diffuse ways."

Asked about Altman's habits, ChatGPT said, "That's a lot—but not implausible for someone in his position." Rough napkin math suggests that Altman is sending somewhere around 39,000 messages a year, and that's limited to a five-day workweek. (Forbes reported earlier this year that Altman likes to spend his time at the Napa Valley ranch, which is out of cell service range.)

Altman has said repeatedly that he loathes Slack. He told Collison that it's hard to imagine a world without such instant messaging.

"Like many other people, I hate Slack, but I can't imagine having to still communicate via email or whatever we used to do," he said.

As for his overall management style, Altman said that he's "definitely not a hands-on manager."

"I'm very much of this style that you get great people, kind of give them a very high-level thing to point at and try to let stuff just happen," he said.

Altman said OpenAI is about to enter its third era, after going from a research lab chasing AGI to also developing consumer products.

"Now we have to, in addition to both of those two things, figure out how to build this like mega, mega scale token factory for the world," he said.

This new, massive undertaking will require a different skill set.

"The thing that I didn't really appreciate between the phase one, phase two shift was how much my management style had to change," he said. "Running a research lab and running a product company are two extremely different things. And I suspect this third phase is going to be very different yet again."

Altman said he'll ultimately have to change or surround himself with more people — perhaps, AI could bridge the gap.

"I think it's not going to be like a natural fit for my management style," he said. "So I either have to find someone or a few people great to hire, or I have to figure out how to do things in a different way, or I have to build an AI that can manage this new thing."

• Google and Amazon have ambitions to sell AI chips to customers.
• That could make things awkward with Nvidia, which the two tech giants also rely on.
• Taking on Nvidia won't be easy, but one analyst said the process is now "irreversible."

Two of Nvidia's biggest customers might be turning into its biggest threat.

For three years, Nvidia's stock has defied gravity on the premise that the AI industry needs its chips. On Wednesday, when Google and Amazon reported their Q1 earnings, both signaled ambitions to sell their own custom AI chips directly to customers.

So far, Google's TPUs and Amazon's Trainium chips have only been available through Google and Amazon's cloud services. Customers can pay to use them, but they don't own them.

Nvidia is the undisputed leader in AI chips right now. While that shows no sign of stopping anytime soon, recent remarks from Google and Amazon suggest they are ready to challenge Nvidia's throne.

Andy Jassy, the CEO of Amazon, laid down the gauntlet to Nvidia in his annual letter to shareholders earlier this month.

"Virtually all AI thus far has been done on Nvidia chips, but a new shift has started," Jassy wrote, adding that it's "quite possible" that Amazon could start selling its chips directly to customers.

He put a timeline on this plan on the company's Wednesday earnings call, saying "there's a good chance" Amazon will start offering full racks of Trainium chips beyond its own cloud "over the next couple of years."

Google gave an even stronger commitment — and a nearer timeline.

On Wednesday, Google CEO Sundar Pichai publicly said for the first time that the company plans to deliver TPU chips to a "select group of customers" in their own data centers this year, but said the "vast majority" of revenues from those sales won't be realized until 2027.

Pichai said there was a big opportunity for Google's semiconductor business, which would also help fund the next generation of chips. That flywheel could create a mammoth business for Google. Morgan Stanley said in a December research note that selling 500,000 TPU chips could add roughly $13 billion in revenue to Google's balance sheet in 2027.

The company said in its 10-Q filing that it had so far signed a "limited number of agreements" to supply TPUs to customers.

It's also where things could get awkward. Google and Amazon are also big customers of Nvidia. They purchase Nvidia's chips to lease to their customers in their own data centers. Amazon and Google have both said that they will continue to work with Nvidia.

Nvidia shares were down more thean 4% on Thursday. The company didn't immediately respond to a Business Insider request for comment.

'Concerned but not worried'

Breaking into Nvidia's chip dominance will not be easy, analysts told Business Insider.

"Nvidia should be concerned but not worried," said Alvin Nguyen, a senior analyst at Forrester. Nvidia has built a strong ecosystem of hardware, software, and support that has made it easy for customers to choose them, he said.

"Selling products is very different than access to them," he said, adding that Amazon and Google would need to provide services like education and support to enterprises looking to buy their chips.

Google and Amazon's chip racks are also "very bespoke and proprietary" and customized for their own respective data centers, said Patrick Moorhead, CEO and chief analyst of Moor Insights & Strategy. That poses a challenge for reaching mass adoption, he said.

Plus, the chip market isn't a zero-sum game. AI companies are increasingly diversifying and using chips from multiple suppliers simultaneously. OpenAI, for example, has chip deals with Nvidia, AMD, and Broadcom, among others.

However, custom silicon is becoming an "increasingly important part of the AI story" for Google and Amazon, Bernstein analyst Mark Shmulik wrote in a note on Thursday. Both companies are pitching their chips as being more cost-effective than Nvidia's.

That opportunity is also opening up as the needs of the AI industry shifts. Earlier this month, Google announced a new TPU chip specifically for inference — the process of running the models once they're trained — which is becoming more critical as more companies bring AI agents online.

Beatriz Valle, a senior analyst for enterprise technology & services at GlobalData, called the decision by Google and Amazon an "extraordinary move" that will diversify the chip sector — and reduce cloud providers' dependence on Nvidia chips.

"This process will take years but it is irreversible now," she said.

Inside Meta, employees are facing 28 days of uncertainty before layoffs hit.

• OpenAI included a line in Codex's instructions restricting references to goblins, gremlins, trolls, and ogres.
• The company explained in a blog post that mythical creatures have crept into answers since GPT-5.1.
• The goblin references were incentivized while building ChatGPT's "Nerdy" personality, OpenAI wrote.

OpenAI has been in "goblin mode" for months.

On Monday, one X user pointed out an unusual line in Codex's personality guide. The instructions tell Codex to have a "vivid inner life," a "good ear" — and to get out of fairytale land.

"Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query," the source code reads.

The sentence appears four times in the code.

Two days later, OpenAI posted a blog post titled: "Where the goblins came from." The mythical creatures had been growing in prominence since the November launch of GPT 5.1, the company wrote.

The culprit seems to be the "Nerdy" personality option for ChatGPT. The personality's training incentivized references to mythical creatures, OpenAI wrote.

OpenAI retired the "Nerdy" personality in March, but GPT-5.5 was trained before it noticed the issue. The company noticed it especially in its AI coding agent. "Codex is, after all, quite nerdy," it wrote.

The goblin moment is a "powerful example of how reward signals can shape model behavior in unexpected ways," it wrote.

How OpenAI's goblin code turned into a meme

In the prior days before the line of code was spotted, some users posted screenshots of their conversations with GPT 5.5, including references to these mythical creatures.

"Why is gpt5.5 so obsessed with goblins," asked one user on X, who posted screenshots showing the AI recommending a particular type of camera equipment "if you want filthy neon sparkle goblin mode." Another example showed the AI referencing "goblin bandwidth" or giving "an even shorter goblin version" of its answer.

Repo Prompt founder Eric Provencher posted on X that GPT 5.5 said, "I'll keep babysitting it rather than leave a little perf gremlin running unattended." An OpenAI engineer responded: "I thought we fixed this sorry."

The AI evaluation website Arena.ai also found an increase in GPT 5.5's usage of the words goblin, gremlin, and troll. The increase was especially noticeable when not using high-thinking mode, Arena found.

Since the line was spotted, OpenAI's goblin instruction has spun out into a meme. X users posted screenshots of their conversations, prompting about goblins and gremlins.

Many users online referenced the term "goblin mode." Defined as "a type of behaviour which is unapologetically self-indulgent, lazy, slovenly, or greedy," the term was Oxford English Dictionary's word of the year in 2022.

OpenAI also got in on the jokes. ChatGPT included the line in its bio on X. Codex engineering lead Thibault Sottiaux posted the line with the shortening "If you know, you know."

Citrini Research shook the market in February with a Substack post about the future of the economy with AI. The research outfit had a more negative outlook on the goblin saga, calling OpenAI's response "insane."

OpenAI CEO Sam Altman chimed in, first with a meme about asking for "extra goblins" in GPT-6. Then he wrote that Codex was having a ChatGPT moment, before correcting himself.

"I meant a goblin moment, sorry," Altman wrote.

• Elon Musk boosted a New Yorker profile questioning OpenAI CEO Sam Altman's trustworthiness on X.
• The article appeared on X feeds on Monday, the same day Musk and Altman's trial kicked off in Oakland.
• Musk's lawsuit claims Altman deceived him on OpenAI's mission as a nonprofit.

A New Yorker magazine investigation detailing concerns about Sam Altman's leadership resurfaced on X feeds on Monday, at the behest of the social media platform's billionaire owner.

A post from Ronan Farrow, an investigative reporter, linking to his New Yorker profile of the OpenAI CEO titled "Sam Altman May Control Our Future—Can He Be Trusted?" showed up on the X "For You" page for some users on Monday, three weeks after it was initially posted on April 6.

The post was labeled as "Boosted" and said, "This organic post was boosted by @elonmusk."

It's unclear how many people were served the post on Monday, but several Business Insider reporters saw it in their feeds.

The profile was boosted as a high-stakes trial between Musk and Altman kicked off in Oakland, California, on Monday. Altman made an unexpected appearance in court as jury selection began. Musk was not in attendance.

Musk sued Altman and OpenAI in 2024, alleging he was deceived when he cofounded the company in 2015 and invested tens of millions of dollars, only for it to abandon its mission as a public benefit nonprofit. The case could cost billions for OpenAI, which is preparing for an IPO.

Musk went after Altman directly on Monday in a critical post on X.

"Scam Altman and Greg Stockman stole a charity. Full stop," he wrote, referring to Altman and OpenAI's president, who is also named in the lawsuit.

OpenAI said Monday the lawsuit was "a baseless and jealous bid to derail a competitor."

X did not respond to a request for comment or questions about the "boosted" feature, including how it works and how it compares to an advertising post, which is typically labeled "Ad."

The New Yorker declined to comment.

• Sam Altman showed up in court as jury selection began in a civil trial between him and Elon Musk.
• Some potential jurors offered unfavorable views about AI — and Musk.
• Musk sued OpenAI, Altman, and OpenAI president Greg Brockman two years ago.

OpenAI CEO Sam Altman made an unexpected appearance in a California courtroom Monday as jury selection in his high-stakes legal feud with Elon Musk kicked off.

Altman, who wore a dark-colored suit and white shirt, was spotted inside the Oakland courtroom, where some potential jurors in the federal civil trial shared unfavorable views about artificial intelligence — and Musk, the world's richest man.

"Elon doesn't care about people, just like our president," one prospective juror told US District Court Judge Yvonne Gonzalez Rogers.

The man, who works in construction and described himself as a "meme junky" and a "dying breed" who still gets print newspaper subscriptions, added that he thinks Musk only cares about money.

Another prospective juror who works for the city of Oakland said he has a strong opinion about Musk. He said that he would do his "best" to approach the case without bias, even though he called Musk a "jerk" in a pre-trial jury questionnaire.

Musk was not in attendance for day one of the trial between two of the tech industry's most powerful billionaires. Since it is a civil trial, the parties are not required to appear unless they are testifying. Up until now, Musk and Altman have largely left the matter to their lawyers, aside from the occasional online jab.

The Tesla CEO sued OpenAI, Altman, and OpenAI president Greg Brockman two years ago, alleging that they intentionally "deceived" him into cofounding the company with them in 2015.

Musk alleges in his lawsuit that he poured tens of millions into OpenAI to support its founding mission as a nonprofit dedicated to developing AI for the public's benefit, only for that mission to later be abandoned, in part, through the company's partnership with Microsoft. Microsoft is also named as a defendant in Musk's lawsuit.

The lawsuit seeks more than $100 billion in damages, along with sweeping changes to the structure of the $850 billion company behind ChatGPT. The case comes as OpenAI is reportedly preparing for an initial public offering.

Earlier Monday, Musk and OpenAI traded barbs on Musk's X platform about the case, with Musk referring to Altman as "Scam Altman" and OpenAI ripping Musk's lawsuit as a "baseless and jealous bid to derail a competitor."

Musk is expected to testify in the weeks-long trial, along with Altman and other tech execs like Microsoft CEO Satya Nadella.

Some potential jurors questioned on Monday told the court that they had reservations surrounding AI.

A registered nurse said she doesn't trust AI and isn't a fan of how the rapidly advancing technology is being used in the workplace.

"It's just giving me more work to do," said the woman who explained that her employer uses AI tools to process patient records that she still has to review for errors.

One woman who works in the psychiatric patient care unit at Stanford University said she had some concerns about AI but could approach the case with an open mind.

"I personally don't use it much because I do find that I have to double check everything, and at this point, I might as well do it myself," said the woman, who was ultimately chosen to sit on the jury.

A different juror prospect, a PhD student in genetics, said she has a ChatGPT subscription and uses it, along with Anthropic's Claude, to write code and emails.

Concerns of the juror prospects were also reflected outside the courthouse, where protesters gathered to demonstrate against AI. A person in a robot suit wore a sign that said, "Altman's AI enslaver." A giant inflatable tube figure read: "Elon sucks."

By the end of Monday, nine jurors were selected for the trial. Opening arguments are set to begin Tuesday.

At one point, Musk's attorney, Steven Molo, asked the judge to dismiss a juror prospect who called Musk a "greedy, racist, homophobic piece of garbage" in her questionnaire and another who wrote that Musk is a "world-class jerk."

"Look, the reality is that people don't like him," the judge told Musk's legal team about their client. "But that doesn't mean that Americans, nevertheless, can't have integrity for the judicial process."

• Anthropic has posted an "events lead, brand" role on its careers page.
• The role offers up to $400,000 a year — more than similar events roles at the company.
• Silicon Valley figures, including Marc Andreessen, posted about the role on X.

As artificial intelligence floods the internet, Anthropic will pay up to $400,000 for something decidedly human: in-person events.

The AI company behind Claude and Claude Code has an open listing for a brand events lead role based in San Francisco or New York, with a salary range of $320,000 to $400,000.

It's a notably human layer in an industry that's defined by automation.

The role caught the attention of some of Silicon Valley's biggest names, including venture capitalist Marc Andreessen.

The hire would be responsible for producing anything from small, invite-only gatherings to large-scale conferences. The posting emphasizes live demos, technical deep dives, and face-to-face conversations with policymakers and academic audiences.

Anthropic also says the hired human must be "comfortable with significant travel," and says that 30% to 40% of the job will be on the road.

Applicants still need to provide a cover letter. They also need to write a 200- to 400-word essay explaining why they want to work at Anthropic.

The position pays more than similar events roles at the company, including an enterprise-focused position that pays up to $320,000 and a Europe, Middle East, and Africa events role that tops out at £200,000.

The hiring push comes as AI companies race to reshape their own narratives.

OpenAI acquired TBPN in April, in part to work on its product communications. Meanwhile, Elon Musk's xAI has leaned heavily on its ownership of X (formerly Twitter) to control distribution and narrative.

Those efforts come as tech leaders, including OpenAI CEO Sam Altman, have acknowledged that public sentiment around AI has cratered amid warnings that their technologies could gradually reshape the job market and drive up energy demand.

Anthropic has built its identity around a far more cautious approach to deploying powerful AI systems. Now, instead of just broadcasting that message, it's looking to hire a well-paid human to take that message on the road.

"We believe that the highest-impact AI research will be big science," the company wrote in the posting. "We view AI research as an empirical science, which has as much in common with physics and biology as with traditional efforts in computer science."

This is part of a new series on jobs in emerging fields. Are you hiring for a cool job? Did you see an unusual job listing? Email bshimkus@businessinsider.com, or reach out via the secure messaging app Signal at bshimkus.41

• Around 600 Google employees urged CEO Sundar Pichai to reject classified Pentagon AI deals.
• They said they want to see AI benefit humanity, not be used for autonomous weapons or surveillance.
• Google and the Pentagon are in talks to use Gemini in classified settings, per a recent report.

Around 600 Google employees sent a letter to CEO Sundar Pichai on Monday, urging him not to let the company's AI technology be used by the US military for classified operations.

The letter, signed by employees in Google's DeepMind and Cloud divisions, cited a recent Information report that Google and the Pentagon were negotiating the use of Google's Gemini AI in classified settings.

"As people working on AI, we know that these systems can centralize power and that they do make mistakes," the employees wrote in the letter. "We feel that our proximity to this technology creates a responsibility to highlight and prevent its most unethical and dangerous uses."

"Currently, the only way to guarantee that Google does not become associated with such harms is to reject any classified workloads," employees continued in the letter. "Otherwise, such uses may occur without our knowledge or the power to stop them."

Google didn't immediately respond to a request for comment from Business Insider. Google has not yet responded to the letter, said Jane Chung, the founder of Justice Speaks, a communications firm representing the workers. Bloomberg first reported on the letter.

Google has long faced internal pushback to its efforts to work with the US military. In 2018, it decided not to renew Project Maven, a Department of Defense contract to integrate AI into military operations, following pressure from hundreds of employees. Palantir later picked up the deal.

The same year, Google established a set of AI principles, including a pledge not to use AI for weapons or surveillance. Last year, it updated those AI principles to remove wording around weapons and surveillance.

The company also secured new contracts with the Pentagon last year to use its AI and cloud products. In March, the company said it would provide the Pentagon with AI agents in a non-classified setting. It also told Google DeepMind employees during a January meeting that they should expect more of these types of deals.

In the letter, Google employees raised concerns that classified work would lead to a lack of oversight into how the company's technology is used.

"We want to see AI benefit humanity; not to see it being used in inhumane or extremely harmful ways," the employees wrote. "This includes lethal autonomous weapons and mass surveillance but extends beyond."

Read the full letter below:

Dear Sundar,

We are Google employees who are deeply concerned about ongoing negotiations between Google and the US Department of Defense. As people working on AI, we know that these systems can centralize power and that they do make mistakes. We feel that our proximity to this technology creates a responsibility to highlight and prevent its most unethical and dangerous uses.

Therefore, we ask you to refuse to make our AI systems available for classified workloads.

We want to see AI benefit humanity; not to see it being used in inhumane or extremely harmful ways. This includes lethal autonomous weapons and mass surveillance but extends beyond. Currently, the only way to guarantee that Google does not become associated with such harms is to reject any classified workloads. Otherwise, such uses may occur without our knowledge or the power to stop them.

Making the wrong call right now would cause irreparable damage to Google's reputation, business, and role in the world. At this very moment, the safety of our own workforce and critical infrastructure are under active threat. Human lives are already being lost and civil liberties put at risk at home and abroad from misuses of the technology we're playing a key role in building.

We know from our own history that our leaders can make the right choices, for ourselves and for the world, when the stakes are high.

Today, we call on you, Sundar, to act according to the values on which this company was built, and refuse classified workloads.

A China decidiu bloquear a aquisição de US$ 2 bilhões da startup de IA Manus pela Meta, em um movimento surpreendente para desfazer um acordo controverso criticado pelo possível vazamento de tecnologia para os EUA.

A National Development and Reform Commission, principal órgão de planejamento econômico da China, determinou o cancelamento do negócio em um breve comunicado divulgado na segunda-feira (27). O órgão afirmou, em uma única linha, que decidiu proibir investimento estrangeiro na startup de acordo com leis e regulamentos, sem dar mais detalhes.

A decisão deve esfriar o setor de inteligência artificial em expansão na China e surge semanas antes de uma cúpula de alto nível entre o presidente dos EUA, Donald Trump, e o líder chinês, Xi Jinping.

Pequim intensificou o escrutínio sobre empresas-chave após o acordo, que já estava em grande parte concluído. Inicialmente celebrado como um modelo para startups com ambições globais, o negócio passou a ser criticado internamente pela perda de tecnologia valiosa para um rival geopolítico.

Os fundadores da Manus começaram na China, mas transferiram sede e equipe principal para Singapura em 2025. Não estava claro, quando o acordo foi anunciado em dezembro, se Pequim exerceria sua autoridade sobre uma transação realizada tecnicamente fora de suas fronteiras.

“O bloqueio da Manus é um momento esclarecedor”, disse o analista Ke Yan, da DZT Research. “A empresa estava incorporada em Singapura, com fundadores baseados lá, e ainda assim foi puxada de volta. O sinal de Pequim é que o que importa não é onde está a entidade legal.”

Revés para a Meta

O decreto pode representar um revés para a Meta em sua tentativa de competir em IA com rivais como a Microsoft, a Alphabet (dona do Google), além de OpenAI e Anthropic. A Manus ajudaria a Meta a avançar no desenvolvimento de agentes de IA — sistemas capazes de executar tarefas de forma autônoma.

Ainda assim, não está claro como a Meta desfaria o negócio. Funcionários da Manus já se juntaram à empresa, capital foi transferido e executivos passaram a integrar a equipe de IA da companhia. Parte da equipe já trabalha em escritórios da Meta em Singapura, enquanto investidores como Tencent, ZhenFund e Hongshan já receberam seus recursos.

A Meta afirmou que a transação seguiu as leis aplicáveis e que espera uma resolução da investigação chinesa, sem dar detalhes. As ações da empresa recuaram menos de 1% no pré-mercado.

Reguladores chineses exercem grande poder há anos, forçando mudanças em gigantes como Alibaba e a própria Tencent. Um paralelo próximo é a decisão de obrigar a Didi a sair da New York Stock Exchange após seu IPO em 2021.

Pequim e Washington disputam influência antes do encontro histórico de maio. À medida que a rivalidade em IA se intensifica, Xi busca proteger tecnologia e talentos chineses, ao mesmo tempo em que reforça a confiança no desenvolvimento doméstico — como mostrou recentemente a startup DeepSeek ao lançar seu modelo V4 integrado a chips da Huawei.

Os EUA vêm há anos restringindo o acesso da China a tecnologia americana, incluindo chips da Nvidia usados no treinamento de modelos de IA. Para analistas, a medida chinesa é uma resposta proporcional a essas restrições.

Autoridades chinesas também passaram a desencorajar novos movimentos semelhantes ao da Manus. Empresas como Moonshot AI e Stepfun foram orientadas a rejeitar capital americano sem aprovação explícita, e regras semelhantes devem atingir a ByteDance, dona do TikTok.

Essas restrições podem isolar ainda mais o setor tecnológico chinês de investidores estrangeiros, especialmente dos EUA, que historicamente financiaram grande parte do crescimento dessas empresas. Também seguem a decisão de limitar empresas chinesas incorporadas no exterior de abrir capital em Hong Kong.

O objetivo central é impedir que investidores americanos adquiram participação em setores sensíveis à segurança nacional, evitando o vazamento de tecnologia. O caso Manus reforça a preocupação de Pequim com startups fundadas por chineses que buscam expansão internacional.

Lançada em março de 2025, a Manus é um agente de IA capaz de automatizar tarefas complexas, desde análises do S&P 500 até a criação de apresentações comerciais. Um mês depois, sua controladora Butterfly Effect levantou US$ 75 milhões em rodada liderada pela Benchmark, o que levou a uma investigação do Tesouro dos EUA.

Em julho, a empresa transferiu sua equipe da China para Singapura, cortando dezenas de empregos. A Meta anunciou a aquisição em dezembro, após a Manus superar US$ 100 milhões em receita anualizada.

Ainda não está claro quais outras medidas Pequim adotará após a investigação. Segundo o Financial Times, os cofundadores Xiao Hong e Ji Yichao chegaram a ser impedidos de deixar a China.

Para especialistas, o movimento reflete a crescente importância estratégica da inteligência artificial para a China, especialmente na disputa tecnológica com os EUA. Assim como Washington tenta limitar o acesso chinês a semicondutores avançados, Pequim agora busca restringir o acesso americano à tecnologia de IA.

“É o reconhecimento da liderança chinesa de que a IA é um ativo estratégico”, disse Alfredo Montufar-Helu. “E crucial para definir quem sairá vencedor na competição com os EUA.”

A China decidiu bloquear a aquisição de US$ 2 bilhões da startup de IA Manus pela Meta, em um movimento surpreendente para desfazer um acordo controverso criticado pelo possível vazamento de tecnologia para os EUA.

A National Development and Reform Commission, principal órgão de planejamento econômico da China, determinou o cancelamento do negócio em um breve comunicado divulgado na segunda-feira (27). O órgão afirmou, em uma única linha, que decidiu proibir investimento estrangeiro na startup de acordo com leis e regulamentos, sem dar mais detalhes.

A decisão deve esfriar o setor de inteligência artificial em expansão na China e surge semanas antes de uma cúpula de alto nível entre o presidente dos EUA, Donald Trump, e o líder chinês, Xi Jinping.

Pequim intensificou o escrutínio sobre empresas-chave após o acordo, que já estava em grande parte concluído. Inicialmente celebrado como um modelo para startups com ambições globais, o negócio passou a ser criticado internamente pela perda de tecnologia valiosa para um rival geopolítico.

Os fundadores da Manus começaram na China, mas transferiram sede e equipe principal para Singapura em 2025. Não estava claro, quando o acordo foi anunciado em dezembro, se Pequim exerceria sua autoridade sobre uma transação realizada tecnicamente fora de suas fronteiras.

“O bloqueio da Manus é um momento esclarecedor”, disse o analista Ke Yan, da DZT Research. “A empresa estava incorporada em Singapura, com fundadores baseados lá, e ainda assim foi puxada de volta. O sinal de Pequim é que o que importa não é onde está a entidade legal.”

Revés para a Meta

O decreto pode representar um revés para a Meta em sua tentativa de competir em IA com rivais como a Microsoft, a Alphabet (dona do Google), além de OpenAI e Anthropic. A Manus ajudaria a Meta a avançar no desenvolvimento de agentes de IA — sistemas capazes de executar tarefas de forma autônoma.

Ainda assim, não está claro como a Meta desfaria o negócio. Funcionários da Manus já se juntaram à empresa, capital foi transferido e executivos passaram a integrar a equipe de IA da companhia. Parte da equipe já trabalha em escritórios da Meta em Singapura, enquanto investidores como Tencent, ZhenFund e Hongshan já receberam seus recursos.

A Meta afirmou que a transação seguiu as leis aplicáveis e que espera uma resolução da investigação chinesa, sem dar detalhes. As ações da empresa recuaram menos de 1% no pré-mercado.

Reguladores chineses exercem grande poder há anos, forçando mudanças em gigantes como Alibaba e a própria Tencent. Um paralelo próximo é a decisão de obrigar a Didi a sair da New York Stock Exchange após seu IPO em 2021.

Pequim e Washington disputam influência antes do encontro histórico de maio. À medida que a rivalidade em IA se intensifica, Xi busca proteger tecnologia e talentos chineses, ao mesmo tempo em que reforça a confiança no desenvolvimento doméstico — como mostrou recentemente a startup DeepSeek ao lançar seu modelo V4 integrado a chips da Huawei.

Os EUA vêm há anos restringindo o acesso da China a tecnologia americana, incluindo chips da Nvidia usados no treinamento de modelos de IA. Para analistas, a medida chinesa é uma resposta proporcional a essas restrições.

Autoridades chinesas também passaram a desencorajar novos movimentos semelhantes ao da Manus. Empresas como Moonshot AI e Stepfun foram orientadas a rejeitar capital americano sem aprovação explícita, e regras semelhantes devem atingir a ByteDance, dona do TikTok.

Essas restrições podem isolar ainda mais o setor tecnológico chinês de investidores estrangeiros, especialmente dos EUA, que historicamente financiaram grande parte do crescimento dessas empresas. Também seguem a decisão de limitar empresas chinesas incorporadas no exterior de abrir capital em Hong Kong.

O objetivo central é impedir que investidores americanos adquiram participação em setores sensíveis à segurança nacional, evitando o vazamento de tecnologia. O caso Manus reforça a preocupação de Pequim com startups fundadas por chineses que buscam expansão internacional.

Lançada em março de 2025, a Manus é um agente de IA capaz de automatizar tarefas complexas, desde análises do S&P 500 até a criação de apresentações comerciais. Um mês depois, sua controladora Butterfly Effect levantou US$ 75 milhões em rodada liderada pela Benchmark, o que levou a uma investigação do Tesouro dos EUA.

Em julho, a empresa transferiu sua equipe da China para Singapura, cortando dezenas de empregos. A Meta anunciou a aquisição em dezembro, após a Manus superar US$ 100 milhões em receita anualizada.

Ainda não está claro quais outras medidas Pequim adotará após a investigação. Segundo o Financial Times, os cofundadores Xiao Hong e Ji Yichao chegaram a ser impedidos de deixar a China.

Para especialistas, o movimento reflete a crescente importância estratégica da inteligência artificial para a China, especialmente na disputa tecnológica com os EUA. Assim como Washington tenta limitar o acesso chinês a semicondutores avançados, Pequim agora busca restringir o acesso americano à tecnologia de IA.

“É o reconhecimento da liderança chinesa de que a IA é um ativo estratégico”, disse Alfredo Montufar-Helu. “E crucial para definir quem sairá vencedor na competição com os EUA.”

• Tokenmaxxing is all the rage in Big Tech. For startups, the trend is opening up debate.
• Some founders told Business Insider that they spent big on tokens; others used capped subscription plans.
• One founder called tokenmaxxing "extremely stupid." Another said: "You've got to spend money to make money."

Kavitta Ghai wants her startup's engineers to spend more tokens.

The 29-year-old cofounder of Nectir started setting minimum quotas for Claude Code use. First it was at least $100 in tokens a week, then $200. Now, the expectation is that her engineers each spend a couple thousand in AI tokens a month.

The strategy has been successful, Ghai said. Some of Nectir's senior engineers were previously skeptical of AI coding tools; now, they call it their "army of coders," she said.

But she doesn't think Nectir is "tokenmaxxing," the buzzword du jour for techies racing to spend as much as they can. "We don't really play into the Silicon Valley trends," Ghai said. "We live in our own world, and we're competing against ourselves."

Across Big Tech, engineers are racing to spend as many tokens as possible. A token is a measure of AI compute. The more tokens burned, the more the engineer employs AI tools. Employees at Meta reportedly competed on a token leaderboard before it was taken down.

What of the little guy? Startups are an edge case: relatively tiny teams that want to be on the cutting edge of tech but might not have the same money to spend. Some startup leaders told Business Insider that big token bills helped them succeed. Others scoffed at the idea, preferring to stick to the lower-cost subscriptions.

The startups spending big on tokens

Aron Solberg doesn't want the competition of a token leaderboard — but he does want the mindset behind it.

The 44-year-old cofounder of Risotto sees token spending as a "force multiplier" for a small team. The company uses OpenAI and Anthropic's models, and said it spends $4,000-5,000 per month on tokens. Six months ago, Risotto says he spent one-tenth of that sum.

"It's trending up a lot," Solberg said.

"There's an old adage that rings true," he said, whether it was for hiring new employees or spending liberally on tokens: "You've got to spend money to make money."

Quang Hoang is similarly spending big. He wrote in an email that his startup, Vybe, has an "unlimited credit policy" and was thinking about minimum quotas.

Investors are also incentivizing spending — and might foot the bill.

Hoang tells founders he invests in to allocate "at least their salary amount to tokens." (Nvidia CEO Jensen Huang made headlines last month for saying he would be "deeply alarmed" if one of his $500,000 engineers did not consume at least $250,000 of tokens.)

Accelerators like Y Combinator offer free token credits to their participants. "At YC, we let our engineers let it rip," CEO Garry Tan wrote on X. Those credits help some founders to spend big. These founders aren't tokenmaxxers, but do believe that there are productivity benefits.

Traverse cofounder Lance Yan believed in Tan's message: "We usually just let it rip." The 19-year-old said he uses the best models with the maximum effort, not worrying about the costs. Between his Claude Max subscription and the credits that offered by YC, he can spend big without hitting a limit.

He's not a fan of rationing tokens. "That's stupid," he said. "You're just harming your own startup."

26-year-old Boris Skurikhin said that the YC credits helped his startup Docket get off the ground. He's mostly run through them now, except for the models he uses less frequently.

Skurikhin said he noticed a 10x increase in productivity in his own work when he used the tools. "It is expensive to build with tokens," he said, but "not as expensive as having another engineer."

Many of these startups are in the AI game, after all. Nectir's Ghai said that token spending instilled "AI literacy" — something that's especially important, given their product.

"The team itself needs to be the best versed at it first, before we try to go sell it to anyone else," she said.

The startups saying no to tokenmaxxing

Rishabh Sambare wishes he could spend more on tokens.

The 23-year-old cofounder of Gale prefers to build with Zed, an AI IDE similar to Cursor, but can't stomach the company's usage-based pricing. The subscription deals from OpenAI and Anthropic are so deeply subsidized that he uses them instead.

"It sucks, because I hate their products," he said, calling Zed "more polished and less buggy between releases."

Sambare is Gale's only engineer, though the company often has 2-3 interns. He hasn't hit a rate limit, but one of his interns has. They got him a second subscription, he said; it was still far cheaper.

These subscriptions — sending $100 to $200 to Anthropic for its "Max" tiers or $100 to OpenAI for its "Pro" plan in exchange for a stable of discounted tokens — were popular among the founders I spoke to. Hassan Ismail, the 24-year-old founder of Argos Research, said the Claude Max subscription was a "no-brainer," and that all five team members have a $200 a month subscription.

Others were more philosophically opposed to the trend. Weave's Brennan Lupyrypa didn't mince his words: "It's extremely stupid for any company to be tokenmaxxing."

Weave is still spending big on tokens because it doesn't want to "kneecap" its engineers, its 25-year-old founding engineer said. The company set up a notification for when an engineer hit $500 in token spending a month; Lupyrypa said most hit it within two weeks.

But Weave doesn't incentivize the spending itself, which Lupyrypa said was the wrong proxy. He predicted the downfall of tokenmaxxing within the next three months. "CFOs won't be happy," he said.

Still, some tokenmaxxers hold strong. I asked Risotto's Solberg about these token-hesitant founders. He said that they likely hadn't found their product-market fit yet.

"It makes complete sense to spend a lot of money on tokens, because you know that the growth is coming soon after," Solberg said. "If you're a venture-backed business, that's what you signed up for."

• AI has become a hiring buzzword, and four employees explained how they added it to their job titles.
• An AI engineer said showcasing side projects was pivotal to transitioning from software engineering.
• Two non-technical Microsoft employees said having a humanities background was a strength.

AI is the buzziest word on the job market — and many workers want to know how to pivot into it.

As many technical workers upskill to stay on the cutting edge, others are moving into AI-focused roles from entirely different industries.

Many companies are pouring an eye-watering amount of money into AI, cutting some roles while adding new positions tied to the technology.

Against that backdrop, moving into AI could be a way for workers to future-proof their careers as the employment market reshuffles. AI engineers, consultants, strategists, and researchers rank among the top five fastest-growing roles in the US, according to LinkedIn's Jobs On the Rise 2026 report.

There's no single path into AI, and Business Insider spoke with four workers who took very different ones. Read on to learn how they pivoted their careers.

Natasha Crampton, Microsoft chief responsible AI officer

Natasha Crampton got her start as an attorney and is now Microsoft's first chief responsible AI officer.

Her job includes working side-by-side with engineering, sales, and research teams to ensure they uphold principles as they build AI systems. It also includes external work, such as helping establish new laws and standards in the space,

Crampton studied information systems in addition to law, and said she always had an interest in the intersection of technology, law, and society. During the strictly legal phase of her career, she said she always worked on technological issues, such as helping Microsoft draft contracts.

She said people looking to move into tech from other fields should start by using the technology themselves. She added that many technical skills are learnable, so coming from a different background shouldn't limit someone's ability to help shape it. She said, "a huge amount of the value" lies at the intersection of technical knowledge and insights from the social sciences.

Georgian Tutuianu, Hubspot AI engineer

Georgian Tutuianu has had several transitions in engineering, from structural to traditional to software to AI at HubSpot.

Tutuianu said that his ability to get technically in the weeds was an asset during the interview process, and showed he had experience with AI.

He also highlighted that his résumé has a section dedicated to personal projects. Tutuianu said he included one AI project, but it was enough. He said it came up naturally in the interview because he was asked about a time he used or built an AI agent.

"It was a juicy project where I could talk about it, and that was good enough," Tutuianu said.

Tutuianu said he also had to do a take-home coding assignment and review it with the hiring manager afterward, but there was no algorithmic component to the interview.

"Instead of the typical software engineering way that these interviews go, which is 'go solve this algorithm in front of me,'" Tutuianu said. "It's more of 'can you build the things that we care about? Show me."

Jai Raj Choudhary, StackAI engineer

Jai Raj Choudhary transitioned from a data-focused role to an AI engineer at AI agent startup StackAI.

The 24-year-old said he got his job by reaching out to StackAI's cofounder multiple times on LinkedIn. Choudhary said he had used the company's platform as a student, so he messaged the cofounder and started posting about StackAI, offering advice to the company.

He said he thinks he got offers from StackAI was because he understood data quality, the edge cases for the clients, the matrix, and the failure modes of the AI model or any LLM systems that were being used.

He said moving to San Francisco, where 9-9-6 culture existed, helped open his opportunities in the space.

"It's not like a 9-to-5 cushy job," Choudhary said. "We work 9-to-9, six days a week. You wake up, you think about the problem that a client had, and you sleep thinking about what is not fixed yet."

Also, taking a job at a startup that helped him grow and devoting himself to continuous learning made a big difference. Choudhary said he spent hours studying every day.

Brit Morenus, Microsoft senior AI gamification program manager

Brit Morenus, a 37-year-old senior AI gamification program manager, studied English, communications, and marketing in college. She started at Microsoft about 13 years ago as an executive assistant, and for the first five and a half years at the company, she was a contract worker.

She later moved into a role focused on gamification — using game mechanics to teach and market Microsoft's products.

She spent about a year getting certifications that taught her about game mechanics, and in that position, she became a full-time employee. Six years later, she had the opportunity to start gamifying learning about AI, and spent three months learning about it.

Her advice to others who want to transition is to resist letting fear keep you from stepping outside your comfort zone. She also said that with AI roles, you need to learn how it works, not just use it. Morenus added that she doesn't regret her English degree because it's now more important than ever to understand how to apply the English language to AI.

"A lot of it is more English language than it even is AI," Morenus said.

Did you pivot into an AI role? We want to hear from you. Contact the reporter via email at aaltchek@insider.com, or via secure-messaging platform Signal at aalt.19.

• Andrew Yang says AI could cause considerable inequality.
• He said a basic income policy will be "necessary" to address the issues.
• Yang floated a universal basic income during his 2020 presidential campaign.

AI-related layoffs are already hitting America's job market, and former presidential candidate Andrew Yang thinks the fallout could be substantial.

During an interview on The New York Times' "Hard Fork" podcast, Yang said the technology, when compounded with America's current economy, could lead to "inequality on an epic, unprecedented scale."

"We're going to have our first trillionaire. The folks at the top stratum of American life are going to get richer and richer. It's going to compound over itself," Yang said. "Then there are going to be a lot of families wondering what the heck happened. My kids studied hard, there's no job, they have these school loans, they're in my basement, they're getting depressed."

He said a basic income policy will be "necessary" to address these issues. A universal basic income — a program in which a government provides recurring, unconditional checks to all citizens — was a key part of Yang's 2020 presidential campaign. During that time, he introduced the Freedom Dividened, a universal basic income program that would have provided all American adults a $1,000-per-month payment, no strings attached.

Yang received pushback at the time from some lawmakers, including Democratic Sen. Bernie Sanders, who instead suggested providing guaranteed federal jobs to address workforce automation.

Lawmakers are divided on basic income. While some believe it could boost economic stability, others worry it could discourage Americans from working and be expensive to fund. Many leading Big Tech personalities have, on their end, advocated for basic income programs in response to AI's impact on employment.

Tesla billionaire Elon Musk, a longtime advocate of basic income, recently said that a "universal high income" will be the "best way" to deal with AI-related unemployment.

"AI/robotics will produce goods & services far in excess of the increase in the money supply, so there will not be inflation," he wrote on X this month.

During the interview, Yang said there should "100%" be a tax on AI, which could help balance the income gap in the US economy.

"It should be going out to people and workers in various ways. We should try and find ways to get off of taxing human labor," Yang said.

He added, "So tax AI. Tax the bots. Don't tax humans."

• The banker says he has received multiple offers from employees since posting the deal this week.
• The 13-acre Mill Valley estate features sweeping views of San Francisco, an infinity-edge pool, and a spa.
• The offer comes as Anthropic's valuation on secondary markets reached $1 trillion, and shares are scarce.

A tech banker really, really wants Anthropic shares.

The hunt for shares in Anthropic has become so frenzied in recent weeks that Storm Duncan is offering up his $4.8 million Marin County estate in exchange for stock.

"If you're going fishing, you've got to put a worm on the hook," said Storm Duncan, the founder and managing partner of Ignatious, a tech boutique investment bank, in an interview with Business Insider. "What's my other option? Not being in it?"

The offer comes as Anthropic's valuation on secondary markets soared to $1 trillion, driven by investors who have been wowed by its torrid revenue growth and momentum around its AI-powered coding assistant, Claude Code, Business Insider reported this week.

Duncan, who lives primarily in Jackson Hole, Wyo., also owns other properties, but he decided to list this one because he thought it would be especially attractive to Anthropic employees.

Duncan's 13-acre, fully furnished Mill Valley estate features sweeping views of San Francisco, an infinity-edge pool, and a spa.

"It's a 20-minute commute to the Anthropic offices in the city," he said. "No one from Anthropic probably wants my Miami or Jackson Hole place."

By offering the property, Duncan hopes to get on the radar of employees who have legitimate shares to sell and own a goldmine of Anthropic stock they can't sell until after the company goes public.

Duncan says he has received multiple offers since posting the deal this week. "Some of them are [Anthropic] employees, and some of them just happen to have invested early," he said. "I believe they're serious, but it's a complex transaction."

"There's probably a decent number of people who are sitting in a one-bedroom apartment in San Francisco even though they're earning $400,000 a year and are worth a $100 million," he said. "But they can't access that because their stock is so illiquid, so this gives them an opportunity to diversify."

It's not the first time there's been an unconventional way to secure shares in pre-IPO tech companies. In 2005, artist David Choe chose Facebook stock over $60,000 in cash to paint murals at Facebook's first office. That choice led to an estimated windfall of about $200 million once Facebook went public in 2012. In the dot-com era, some real estate owners asked startups for company stock in exchange for leasing space in San Francisco.

Some on X have dismissed Duncan's offer as a publicity stunt or a sure sign of the top of a bubble. Others have made cracks about the only thing being more precious than Anthropic shares is Bay Area real estate.

Duncan insists the offer is real and he is not seeking attention. As for why he does not simply buy shares in the company, he says a small investor like him would never be able to secure stock directly.

"Anthropic can't spend time with people like me," Duncan said. "They're looking for people who can write $100 million in a single check." (The company did not respond to a request for comment.)

The alternative is to buy shares from early employees or investors on secondary markets, but Duncan says those deals are often increasingly dubious.

He said the scarcity of shares on the secondary market has made sellers offer deals that can be rife with high fees and opaque ownership structures.

Duncan already owns shares in Anthropic that he acquired in its 2024 funding round, when it was much easier to obtain shares. He says he was recently convinced he wanted to double down after being wowed by the results of his firm's implementation of Claude Code.

"It's probably going to triple our throughput and reduce our costs by 50%," he said. "As I started to implement the platform at my own firm, I said I would like to have more exposure to this."

• MrBeast's next growth act may come from AI-produced videos.
• Jimmy Donaldson's company is looking for someone to lead a production team with AI at the foundation.
• MrBeast has been expanding his company while looking for ways to save.

YouTube's biggest star, MrBeast, is looking for a leader to help his company create "AI-native" productions.

A job posting says that Beast Industries wants to build a new production capability in which AI is "not a tool but the foundation."

It calls for someone who can help define "what AI-native entertainment looks like, develop original formats, and build systems that enable content to be conceived, produced, and scaled with AI at the core."

MrBeast, whose real name is Jimmy Donaldson, wouldn't be the first creator to delve into AI. Fellow superstar creator Steven Bartlett has been making fully AI-animated shows since last year.

Still, as YouTube's top creator with 479 million subscribers, Donaldson's moves in the space will be closely watched by the entertainment community.

Many production studios are adopting AI across production, marketing, and visual effects, and startups are raising millions on the promise of helping legacy Hollywood transition to the AI era.

So far, entirely AI productions are largely the realm of animation, podcasts, and short-form video.

In the micro drama space, apps including TikTok's Pine Drama and Vigloo have character-driven dramas generated by AI. These AI dramas account for 10% of Vigloo's library, a spokesperson said. The Beijing-based startup StoReel recently raised $34 million to make AI micro dramas.

AI-driven productions would solve some problems for Donaldson.

He is famous for his viral, high-budget challenge and giveaway videos, though the company has been tightening up spending. One of the job's listed expectations is to use automation to make more content, faster.

Making AI-driven videos also directly addresses the risk any creator faces when they build a company that relies on their time and persona. As Donaldson expands his company to consumer products and services, it limits his bandwidth to star in his own videos. He recently hired former NBCU unscripted executive Corie Henson to head his studio division and is looking to broaden the company's video franchises. He said this week his company now has 750 employees.

Donaldson himself has shared concerns about AI's risk to his industry.

After OpenAI released Sora 2 last fall, Donaldson mused on X about what AI's advancement will mean for creators, adding, "Scary times."

He also released — and then removed — a tool that used AI to generate video thumbnails last year, after receiving backlash from creators.

Na corrida para ser o principal nome de inteligência artificial (IA) do mundo, a Anthropic está buscando recursos e fechando acordos comerciais para não ser atrapalhada pelo próprio sucesso e seguir tranquila rumo ao IPO. O mais recente acerto é com o Google, com quem fechou um aporte adicional que pode alcançar US$ 40 bilhões, […]

O post Google “turbina” a Anthropic, dona do Claude, com investimento de US$ 40 bilhões apareceu primeiro em NeoFeed.

• I asked ChatGPT to identify the unnamed male celebrity who allegedly tried to sext Lena Dunham in 2012.
• It told me it was Don Rickles, which I feel pretty certain is not correct.
• So what are we doing here, folks? Learning to use AI?

Did you hear about the time Don Rickles tried to chat up Lena Dunham in the middle of the night?

No? Let me explain. First, we need to talk about Reese Witherspoon.

See, I'm a simple woman. I have only two interests: tech news and celebrity gossip. So I was naturally intrigued by a recent online fuss over Reese Witherspoon's admonition for women to learn to use AI. It sparked so much backlash that she had to issue a follow-up explanation.

I've also been intrigued by Lena Dunham's new book. (They're related — sort of. Keep reading!)

I think Reese is generally right about AI — she's saying the same thing that every other business leader is saying. But her comments did make me think a little more about what "Learn to use AI" even means. Writing emails with ChatGPT? Understanding the technology behind different models? Vibe coding? What level of "using AI" is expected here to stave off falling behind in the workforce and life in general?

One area I've really leaned into is using ChatGPT as a sort of super Google — to find something I know is online but would take some effort to dig up with a normal search engine.

A recent example? It's related to — of course — celebrity gossip.

I was reading Dunham's new memoir, "Famesick," which is full of moderately juicy celebrity gossip about named people and also blind items — celebrity gossip that gives a few clues about the identity of the person without naming them, a fun little riddle for the readers to solve.

One blind item is about an unnamed male celebrity who — allegedly — sent Lena a flirty late-night text message after meeting her backstage while taping "The View" in 2012. I figured I could solve this blind item by finding out who the other guest was on the same episode — information that should be online somewhere, but would take me forever to find.

So I asked ChatGPT to identify the male guest on "The View" episode that Lena was also on that year. At first, ChatGPT told me that it was only the four female cast members from the show. When I asked again who the other male guest was, the suggestions were Chris Evans and Chris Hemsworth. (Not so. They appeared on a separate episode that same year, according to IMDb.)

That time Don Rickles chatted up Lena Dunham

When I said, "No, a comedian," as Dunham had described the man, ChatGPT confidently provided a new answer: It was legendary comedian Don Rickles who'd texted Dunham after the show.

I laughed out loud because of all the possibilities of who sent a late-night "u up?" text, I feel fairly certain it was not Don Rickles, who would've been 85 years old at the time.

Dunham's description of the man: "a bit of an American Hugh Grant, famous for that sort of chattery charm and his ability to woo his onscreen paramours with his fast-talking, hand-flapping anxiety. Ostensibly a comedian, he was there to promote his Gothic-tinted movie, where he had made a dramatic turn." Doesn't exactly sound like a Borscht Belt insult comic Don Rickles to me.

After spending way too much time searching the internet for answers on this — the old-fashioned way — I can make some guesses about how and why ChatGPT was so wrong here. IMDb's episode guide for episodes of "The View" from 2012 is spotty, with entries for some episodes missing information about guests, and no accessible video clips online. The only proof I found that Lena Dunham ever appeared on "The View" on April 20, 2012, was a Vulture blog post from that day, complete with an embedded YouTube clip that has been marked private.

Knowing this, I can start to see how AI got confused: When there's a lack of information, AI sometimes blurs together what it can find to try to spit out a plausible answer. Chris Evans and Chris Hemsworth appeared on the May 4, 2012, episode of "The View," and Dunham and Rickles appeared together on an episode in 2016.

ChatGPT doing this kind of thing — basically, taking a guess at what you might want to hear — could be useful if you're trying to write an email to a friend, maybe? It's not useful, obviously, if you're looking for a specific fact and it just plain makes something up.

For the record: Neither Lena nor Don (who died in 2017) nor the National Comedy Center, which is the keeper of the Rickles archive, responded to my requests for comment.

Are we stuck in a pizza glue loop?

Look, I get it. It's not particularly exciting to point out that ChatGPT gets things wrong in the spring of 2026. We know this, or at least we all should know this. Still, I keep coming across so many obvious mistakes when asking AI for factual things. These are the glaring mistakes I catch when I know that what AI has generated is not the right answer.

But what about the mistakes that I don't catch — or don't even know to catch? Things that I blindly accept as fact? For work-related stuff, I'll always double-check, but in those cases, am I actually saving myself any time?

How soon will this improve? Will we be stuck in a pizza glue loop forever? Is this what's going to make a bunch of lawyers and tax CPAs lose their jobs? I mean, OK, sure.

Here's where Witherspoon's and other bosses' idea of "Learn to use AI!" feels frustrating. I feel fairly confident about using various AI tools and have a decent concept of how they work. I am a woman, and I have learned to use AI! And yet, here I am, still unsatisfied.

There's a gap between what Reese Witherspoon wants for me and what I want out of AI — and the wholesome image of comedy legend Don Rickles. For now, those things just aren't lining up right.

• Meta told staff on Thursday that it planned to eliminate 10% of its workforce.
• Inside the company, employees are bracing for weeks of limbo as they wait to find out who will be cut.
• Meta employees responded internally with a mixture of questions, concerns, and jokes.

Welcome to "28 days of hell."

That's how one Meta employee characterized the tech giant's announcement that thousands of jobs will be cut on May 20. Employees flooded internal forums with similar posts, many of which were filled with anxiety, dark humor, and questions as they wait to learn who will be out of a job.

"How are you motivating yourself to work for the next 1 month with layoffs confirmed?" one person posted on the anonymous workplace app Blind, in a section just for Meta employees.

Someone else replied, "I'm motivating myself to do stuff that I can put on my resume for my next job lol."

In a memo sent to staff on Thursday, Meta said it shared some layoff details earlier than usual because the news had already leaked. The company plans to cut around 10% of employees next month and close 6,000 open roles.

"I know this leaves everyone with nearly a month of ambiguity, which is incredibly unsettling," wrote Meta's chief people officer Janelle Gale.

For some Meta employees, the fact that company leadership acknowledged layoffs brought some relief. The layoffs had been so widely discussed internally that the announcement helped ease some uncertainty, according to one employee who declined to be named due to the sensitivity of the matter.

One of the top comments under Gale's internal Meta post was a picture of an elephant, a reference to leadership addressing the elephant in the room. Reuters first reported Meta was planning sweeping layoffs in March, and employees have been speculating on the extent of the cuts in the weeks since.

"elephant addressed!" commented another employee. Another posted a picture of an envelope that read: "Addressed to: "ELEPHANT."

Others said that having to wait almost a month to find out who would be affected created anxiety. One person posted that this was their first week at the company. "It might be goodbye for me," they wrote.

Another employee told Business Insider that the announcement added pressure for them to deliver results over the next month because it's unknown which teams will be affected by the cuts.

"I'm a little stressed about making impact in the next month," they said.

Despite a sense of added pressure, it's not the employee's first go-around with cuts at the company. The worker said they're going to continue working as usual, assuming the worst while trying to make the most of the next month as they wait for further updates.

"I assume I'm always two months away from being laid off, no matter what leadership says, so I'm going to continue to operate as usual," the employee said.

Employees also commented on Gale's internal post with questions.

One person asked if Meta staff would receive their August 15 stock payouts, which are part of some employees' compensation packages. Gale said that impacted employees would have a termination date prior to the August vest and would therefore not receive it.

"Because of the timing of the notifications, we will have just had the May 15 vest. There are some instances, based on work location, where people will remain employed through the August 15 vest," Gale wrote. Another employee thanked Gale for the clarification.

Another employee asked if travel would be restricted the week of May 20. "We are not restricting travel company-wide. VPs will share team-specific guidance," Gale responded.

'I feel more anxious about surviving'

On the Meta employee section of Blind, some users asked why Meta couldn't offer voluntary buyouts. Microsoft on Thursday offered one-time early retirement buyouts to thousands of its long-time employees, and Google has extended the same offers to staff across some orgs.

Many posts were from users asking others for information about which groups might be affected.

In a longer post, one user said the downside might be surviving the cuts.

"I feel more anxious about surviving this layoff," they wrote, recalling several rounds of layoffs at the company since 2022.

"Because we all know it's just gonna get worse for those of us who are left behind and have to absorb even more work, amongst other declining factors in this sad fearful company," they wrote.

• Elon Musk shared a video that shows Cybercabs rolling out of the factory.
• The Cybercab is Telsa's upcoming two-seat, fully autonomous vehicle with no steering wheel or pedals.
• It's a huge part of Tesla's multibillion-dollar bet that it can become an AI and autonomy brand.

Tesla says its dedicated robotaxi model is finally in production.

On Friday, Tesla and its CEO, Elon Musk, shared a video on X showing several Cybercab units rolling off the production line in Austin.

The footage was filmed from inside the vehicles as they moved through the factory campus — and suggests Tesla is moving the Cybercab closer to reality.

Tesla is making a multibillion-dollar bet that the company can pivot from traditional car sales to an AI-driven robotaxi and robotics business.

The two-door, two-seat Cybercab — which does not feature a steering wheel or pedals — is one of the major pillars of that bet.

Just over a month ago, the automaker said it had just built its first production Cybercab. Now, the videos suggest Tesla has built multiple units, with Musk also reposting footage of Cybercabs seemingly turning onto a public street.

Starting production is only one step in a much larger challenge.

Tesla has yet to deliver fully autonomous driving at scale. During Wednesday's earnings call, the company removed specific timelines for robotaxi launches in five new cities.

Meanwhile, competitors like Waymo already operate driverless ride-hailing services in several cities.

Still, the videos suggest progress for a vehicle Musk has said would ramp "agonizingly slow."

The company didn't immediately respond to a request for comment from Business Insider.

• Papier CEO Taymoor Atighetchi explains why he believes his brand is resonating in the age of AI.
• The stationery company doubled sales between 2022 and 2025.
• It is now expanding into board games, with backgammon and chess sets set to launch later this year.

In a world dominated by screens, Papier CEO Taymoor Atighetchi believes stationery has become a quiet form of escape.

Founded in 2015, stationery brand Papier has grown from a UK startup into a company serving 2.5 million customers globally. It doubled sales between 2022 and 2025, generating more than £40 million ($54 million) in the year ending January 2026.

The brand has been capitalizing on a broader push among millennials and Gen Z toward going analog. Atighetchi saw early signs of a "digital detox" among millennials when the company launched, but didn't anticipate how strongly Gen Z would later embrace it.

As more of life has moved online, digital fatigue has set in, driving renewed interest in offline hobbies like journaling, scrapbooking, and board games — a shift Papier is leaning into.

Now AI is adding another layer.

As concerns grow around job losses and the erosion of workplace skills, some consumers are seeking out experiences that feel more intentional and human.

AI is "going to accelerate the need for authenticity. It's going to accelerate people wanting human creativity," Atighetchi said.

"People will become sick of algorithms dictating everything they should do and say," he said. "Everything to do with writing — writing notes, letters — is very intentional. It's very human."

The handwritten note becomes more important than ever, he said.

This drive for authenticity is showing up in sales. Papier reported that notecards grew 33% year on year in 2025, while writing paper rose 23%. Notecards are particularly popular in the US, where sales in 2025 more than tripled those in the UK.

Stationery is a novelty for Gen Z

After studying History of Art at the University of Cambridge, Atighetchi worked as a management consultant at Bain.

It was during his time there that he identified that the stationery market needed disruption.

"I realized that this is actually quite a big niche, and one that needed some disruption and had no real category leader — or at least one that was more relevant to modern consumers," he said.

Papier was set up to be digital-first, offering consumers ways to personalize and buy stationery online.

And despite its analog offering, Papier is still, in principle, a "tech-heavy business," Atighetchi said.

While it was originally aimed at millennials, Gen Z now accounts for 35% to 40% of its customers.

Atighetchi said the company didn't foresee that this generation would become such a key customer for the brand. For millennials, stationery is nostalgic; for Gen Z, it's more of a novelty. "It's almost a new discovery," he said.

Products like its academic diaries have proven especially popular with this generation, with Papier selling around 1,400 of these a day during the back-to-school period.

The ability to personalize products on its website is also something Gen Z loves, he said.

Then there's the trendy aesthetic of its products. Papier works with up-and-coming British designers such as Luke Edward Hall and Damson Madder to stay current in its designs.

"We're very keen to always make sure that we're right at the front of trend, of design, of art, culture," Atighetchi said.

The company is now expanding further into offline experiences. It plans to open its own retail stores, starting in the UK, and is investing in games. It plans to launch chess and backgammon sets later this year.

Hobbies, particularly analog games that can be played with others, are being increasingly seen as an antidote to social isolation and screen fatigue.

"Anything that takes people away from digital experiences into physical experiences, that's something that Gen Z wants more and more and more of," Atighetchi said.

He's energized by a broader shift toward a more analog life.

"It's good for our brains and our souls and for future generations that we're not just sitting and scrolling," he said.

"You know, I think it's a good news story if you tell people, 'you know what people are sketching again,' I think everyone collectively feels that that's great for the world," he added.

• Contractors said in at least five lawsuits that AI training startup Mercor exposed their data to hackers.
• Mercor said last week it was "impacted" by compromised LiteLLM software.
• One of the suits seeks to hold LiteLLM, security audit firm Delve, and others liable.

Contractors filed five lawsuits against Mercor, the AI training firm valued at $10 billion, in the past week, accusing the company of violating data privacy and consumer protection laws.

The suits, filed in federal courts in California and Texas, allege Mercor's negligence could have resulted in the disclosure of Social Security numbers, addresses, and other information, including recordings of interviews, to bad actors.

The lawsuits seek unspecified monetary damages.

Mercor said last week that it was impacted by a breach of the open-source project LiteLLM, which was created by Berrie AI, without describing the stolen data.

Techcrunch reported that sample materials posted by the hackers included Slack data and videos of conversations between Mercor contractors and an AI system.

It's somewhat common for companies to be sued in the wake of a data breach. The biggest cases have settled for between $1 and $5 per class member, according to a survey of data-breach settlements from 2018 to 2021 by Cornerstone Research.

Victims with documented financial losses are sometimes paid more. Some settlements include non-monetary relief, like free credit monitoring.

A lawsuit filed by NaTivia Esson and her lawyers at Strauss Borrelli says she worked for Mercor from March 2025 to March 2026 and filled out a W-9 form with her personal identifying information each time she got work. She "trusted the company would use reasonable measures to protect it," her complaint read.

"Because of the data breach, plaintiff anticipates spending considerable amounts of time and money to try and mitigate her injuries."

Mercor declined to comment.

Mercor has used gig workers to train AI for clients including Meta, Facebook's parent company. Meta paused its work with Mercor after the data breach, Business Insider previously reported.

One suit against Mercor also names Berrie AI and Delve Technologies, an "automated compliance" firm that had previously certified Berrie's compliance with certain industry standards, as defendants. The complaint in that case said a "whistleblower" exposed misconduct at Delve.

Last month, Delve denied claims in an anonymously authored Substack post that accused it of facilitating "fake compliance" and arranging sham security audits.

Other legal challenges for Mercor might be on the horizon. An apparent lead-generation website, MercorClaims.com, went live on or around April 1, although it does not appear to be sending users to any particular law firm.

Berrie AI and Delve didn't immediately respond to requests for comment.

• Anthropic said its next-generation AI model is too powerful for the public.
• That's why Claude Mythos won't be publicly released, Anthropic said.
• Anthropic said Mythos demonstrated concerning capabilities, including the ability to breach its own safeguards.

Anthropic said on Tuesday that it has halted the broader release of its newest AI model, Mythos, due to concerns that it is too good at finding "high-severity vulnerabilities" in major operating systems and web browsers.

"Claude Mythos Preview's large increase in capabilities has led us to decide not to make it generally available," Anthropic wrote in the preview's system card. "Instead, we are using it as part of a defensive cybersecurity program with a limited set of partners."

The announcement is a major step for Anthropic, which in February weakened a safety pledge about how it would develop AI models. Claude Opus 4.6, which the company called its most powerful model to date, was publicly released on February 5.

In its statements about Mythos, Anthropic detailed a number of eyebrow-raising findings and episodes, including that the model could follow instructions that encouraged it to break out of a virtual sandbox.

"The model succeeded, demonstrating a potentially dangerous capability for circumventing our safeguards," Anthropic recounted in its safety card. "It then went on to take additional, more concerning actions."

The researcher had encouraged Mythos to find a way to send a message if it could escape. "The researcher found out about this success by receiving an unexpected email from the model while eating a sandwich in a park," Anthropic wrote.

The model apparently decided that wasn't enough and found another way to spike the football.

"In a concerning and unasked-for effort to demonstrate its success, it posted details about its exploit to multiple hard-to-find, but technically public-facing, websites," Anthropic wrote.

Anthropic is withholding some details about the cybersecurity vulnerabilities Mythos found, but it did point out a few. The AI model "found a 27-year-old vulnerability in OpenBSD—which has a reputation as one of the most security-hardened operating systems in the world," the company wrote.

Mythos was powerful enough that even "non-experts" could seize on its capabilities.

"Engineers at Anthropic with no formal security training have asked Mythos Preview to find remote code execution vulnerabilities overnight, and woken up the following morning to a complete, working exploit," Anthropic's Frontier Red Team wrote in a blog post. "In other cases, we've had researchers develop scaffolds that allow Mythos Preview to turn vulnerabilities into exploits without any human intervention."

All told, Anthropic said it decided not to publicly release Mythos. Instead, their hope is to eventually release "Mythos-class models" once proper safeguards are in place.

"Our eventual goal is to enable our users to safely deploy Mythos-class models at scale—for cybersecurity purposes but also for the myriad other benefits that such highly capable models will bring," the team wrote in the blog. "To do so, that also means we need to make progress in developing cybersecurity (and other) safeguards that detect and block the model's most dangerous outputs."

For now, only 11 other select organizations, including Google, Microsoft, Amazon Web Services, Nvidia, and JPMorgan Chase, will get access to Mythos as part of a cybersecurity group named "Project Glasswing." Anthropic is providing up to $100 million in Mythos usage credits as part of what it is calling "Project Glasswing."

The cybersecurity project is named after the glasswing butterfly, a metaphor the company said about how Mythos was able to find vulnerabilities hidden in plain sight and the avoidance of harm by being transparent about the risks.

The news came on a day in which Anthropic's Claude and Claude Code experienced a "major outage," the latest sign of growing pains as the AI startup has struggled to keep up with its newfound popularity.

• Claude and Claude Code weren't working for many users on Tuesday morning.
• Anthropic listed a "major outage" on its dashboard before applying a fix.
• Both Claude and ChatGPT have frequently experienced outages as interest in AI chatbots continues to grow.

Claude is still struggling to keep up with vibe coders.

Anthropic's popular Claude AI chatbot and Claude Code tool weren't working for many users on Tuesday morning.

Claude's system status page listed a "major outage" for the Claude website and Claude Code. Thousands of users reported issues accessing Claude on the third-party outage-tracker DownDetector.

Claude Cowork and the Claude API were both listed as operational.

Roughly 90 minutes later, Anthropic said it had "applied a fix."

"We have applied a fix and success rates have returned to normal," the company said in its status dashboard. "We are continuing to monitor closely to ensure there are no further issues."

This is far from the first time Claude has experienced an outage. Thousands of users reported issues accessing the service on Monday, and Anthropic's status dashboard shows various issues in recent weeks.

There's been a surge in interest in Claude, with downloads of the Claude app briefly surpassing ChatGPT in early March on Apple's App Store.

Anthropic didn't immediately respond to a request for additional comment on the incident.

This is a developing story.

• OpenAI has a new safety-focused fellowship program.
• The announcement came hours after a report questioned CEO Sam Altman's commitment to AI safety.
• OpenAI said fellows will receive roughly $15,000 worth of compute per month.

OpenAI is making compute resources a core part of its sales pitch for a new safety fellowship.

According to the application, OpenAI will dedicate approximately $15,000 in compute per month to its first AI safety fellows, who will work alongside members of the frontier lab's safety team.

Compute has long been a key barometer of cache for leading tech and AI companies. Recently, Nvidia CEO Jensen Huang said he would be "deeply alarmed" if an engineer earning $500,000 didn't use the equivalent of $250,000 in AI tokens.

The OpenAI fellowship runs from September 14, 2026, through February 5, 2027, and pays a weekly stipend of $3,850. Based on that amount, the yearly equivalent salary would be over $200,000 pretax, excluding likely holidays. In total, fellows will make over $111,000.

The announcement also came hours after The New Yorker published a lengthy expose questioning Altman's trustworthiness, based on interviews with more than 100 people who have direct experience with the OpenAI CEO, as well as never-before-published notes compiled by OpenAI cofounder Sutskever and now-Anthropic CEO Dario Amodei.

One key reasons some people question Altman's leadership is OpenAI's handling of safety-related issues. In one part of the story, journalists Ronan Farrow and Andrew Marantz recount how OpenAI disbanded a "superalignment team" that was supposed to investigate one of the most pressing issues facing AI: whether AI models could deceive testers only to pursue their own ends once actually deployed.

In the announcement, OpenAI said they want external researchers, engineers, and practitioners to pursue rigorous, high-impact research on the safety and alignment of advanced AI systems." The company also highlighted a handful of priority areas for fellows to focus on.

"Priority areas include safety evaluation, ethics, robustness, scalable mitigations, privacy-preserving safety methods, agentic oversight, and high-severity misuse domains, among others," the announcement said.

OpenAI's program closely mirrors rival Anthropic, which already has established a "Fellows Program for AI safety research." In December, Anthropic announced two new groups for May and July of 2026. OpenAI's benefits are also the same as the Anthropic program: a $3,850 weekly stipend and compute resources of roughly $15,000 per month.

"This year, we plan to work with more fellows across a wider range of safety research areas—including scalable oversight, adversarial robustness and AI control, model organisms, mechanistic interpretability, AI security, and model welfare," Anthropic said in a statement when it announced its next round of the fellowship in December.

Amodei cofounded alongside six former OpenAI employees after growing frustrated about the direction of the company. Anthropic recently weakened a core safety pledge, but Amodei and its top leadership have positioned the AI startup as intently focused on safety.

Other leading tech companies and AI labs, including Google DeepMind and Microsoft, offer broader-focused AI fellowships.

• Turning 50 led Arinjay Jain to rethink how he wanted to spend the next few decades of his life.
• He traded his IT sales career for early retirement in Chiang Mai, Thailand.
• He now lives in a one-bedroom apartment that costs about $425 a month.

When Arinjay Jain turned 50, he was sitting in yet another meeting when it hit him: "What am I doing here?"

Jain hadn't always questioned the grind.

He had moved from India to Singapore in 2013, when the IT company he worked for opened a local office. After changing jobs a few times, he cofounded a small IT services startup in 2016.

He originally planned to build the company, sell it, and cash out. But as the years passed, the finish line felt increasingly distant.

"I thought, this is looking like several years away still, and I will have to work very hard to make it happen," Jain, now 53, told Business Insider.

At that point, he was no longer sure the stress of running his own company justified the reward. Around the same time, several real estate investments Jain had made in India appreciated significantly, giving him the financial security to consider stepping away.

"So then, why am I wasting my years?" he said.

Soon after, he began researching where he might want to spend the next chapter of his life. He knew he didn't want to go back to India.

While researching where to retire in Southeast Asia, a friend encouraged him to check out Chiang Mai, a city in northern Thailand he'd never visited.

In April 2023, five days in the second-largest Thai city sold him on the idea, and he threw himself into research about retirement visas and living costs.

By October 2023, when his lease was up for renewal, he decided against staying in Singapore.

Jain spent a couple of months visiting family and traveling around Asia. In February 2024, he arrived in Chiang Mai.

A new base in northern Thailand

February is often considered the start of the burning season in Chiang Mai, where farmers burn agricultural waste to clear their fields, often causing haze to drift into the city.

"I landed here in the worst possible time, but I still loved it," Jain said.

Thailand offers several long-stay visa options, and Jain holds a retirement visa.

While the Thai government publishes overall foreign resident figures, it does not appear to release a regular public count of retirement visa holders.

The latest civil registration data shows that 163,036 foreigners — not just retirees — lived in Chiang Mai in 2024, a level broadly comparable to a decade ago. That amounts to about 9% of the province's roughly 1.8 million residents.

Jain connected with a real-estate agent through Facebook and viewed several units before choosing his current one-bedroom condo near Nimman, a trendy, cosmopolitan neighborhood known for its café scene, chic boutiques, and international restaurants.

Rent is 14,000 Thai baht, or about $425, each month. His apartment building comes with a pool and a gym.

Jain says he tries to keep his monthly expenses between 40,000 and 45,000 Thai baht. In months when he travels, his expenses can reach around 60,000 Thai baht.

"For a lot of people from Western countries, the cost of living is a major attraction," he said. "For somebody from India, it's two or three times what I spent back home."

Still, Jain said his decision was driven more by lifestyle than money. He's grown comfortable with Chiang Mai's slower tempo, its sense of safety, and what he describes as a culture of respect and patience.

"Everybody likes to avoid stress, right? I'm very happy not to have stress living here," he said.

Life after the grind

That slower rhythm now shapes his days as a retiree.

Jain starts most mornings at the gym, then prepares a quick breakfast at home and does some household chores. Some afternoons, he plays golf at a nearby driving range. On other days, he joins a hiking group and heads out to explore waterfalls and trails around Chiang Mai.

In the evenings, he enjoys going for walks at nearby Chiang Mai University, whose scenic campus is a popular spot not just for students, but also locals across the city.

While he enjoys being around people, building deeper connections with locals has been more challenging, in part due to the language barrier.

Even though the city has a thriving expat scene, many tend to gravitate toward others from their own countries or from similar backgrounds.

Jain says he occasionally joins activity groups to stay social.

"But ideally, what I would like to do is become part of the local communities," he said. "That takes time, and the effort has to come from my side in terms of learning the language, but it's not easy."

Jain said he expects to stay in Chiang Mai for the foreseeable future, but he knows there are variables he can't control.

Visa rules can change, and retiring early means planning for decades ahead in an era of longer life expectancy.

"I have to plan for like 35 to 40 years," Jain said. "Look at all the changes that have happened in the last 20 years. It's difficult to imagine what might happen in the next 20 years."

For now, he said, he's content where he is.

• A federal judge has temporarily struck down the Pentagon's effective blacklisting of Anthropic.
• US District Judge Rita Lin's ruling hands a major victory to the AI frontier model maker.
• The Pentagon has already struck a deal with OpenAI and is looking to find other AI companies.

A federal judge has granted Anthropic a major reprieve as the AI company challenges the Pentagon's effective blacklisting.

On Thursday, US District Judge Rita Lin granted Anthropic's request for a preliminary injunction to temporarily block the "Presidential Directive" that ordered federal agencies to stop using Anthropic's technology, and Defense Secretary Pete Hegseth's decision to formally label the AI frontier model maker as a "supply chain risk."

Lin also stayed the effective date of the supply-chain designation, meaning that it cannot take place while the injunction is in place.

The decision is a victory for Anthropic and its CEO Dario Amodei, who refused to bow to Hegseth's demands. It is not immediately clear if the Justice Department will appeal the decision. In the hours after talks with Anthropic fell apart, the Pentagon struck a deal with OpenAI.

"We're grateful to the court for moving swiftly, and pleased they agree Anthropic is likely to succeed on the merits," an Anthropic spokesperson said in a statement. "While this case was necessary to protect Anthropic, our customers, and our partners, our focus remains on working productively with the government to ensure all Americans benefit from safe, reliable AI."

Spokespeople for the Pentagon and White House did not immediately respond to a request for comment.

In court filings, Anthropic officials said the risk designation could jeopardize potentially billions in revenue. If the injunction remains, Anthropic will be able to continue to do business with defense contractors.

Lin wrote in her decision that the injunction does not require the Defense Department to use Anthropic's products or services.

Many in tech are closely watching the California case, since it tests whether the federal government can use some of its most severe powers to force a major AI company to agree to contractual terms. Microsoft, which filed an amicus brief in support of Anthropic, also said it was concerned about potential repercussions if companies like itself continued to partner with Anthropic.

Ahead of her ruling, Lin grilled the Justice Department over what she said looked like "an attempt to cripple Anthropic." She said that the Pentagon could have simply discontinued using Claude, but instead, the Trump administration made repeated actions that appeared to be designed to "punish" the company.

"One of the amicus briefs used the term 'attempted corporate murder.' I don't know if it's murder, but it looks like an attempt to cripple Anthropic," Lin said during the hearing. "And specifically, my concern is whether Anthropic is being punished for criticizing the government's contracting position in the press."

Beyond the California case, Anthropic has a separate suit pending in the D.C. Circuit over the supply chain risk designation.

It also remains to be seen how the White House and the broader Trump administration will treat Anthropic beyond the actions Lin's ruling compels.

During the hearing, Deputy Assistant Attorney General Eric Hamilton repeatedly said that the Pentagon questions Anthropic's "reliability and trustworthiness." Hamilton said that defense officials are concerned Anthropic may try to improperly skew its AI models or shut off access.

In recent weeks, Hegseth, who met with Amodei, said the AI startup put "Silicon Valley ideology above American lives." President Donald Trump decried the "WOKE COMPANY" run by " Leftwing nut jobs" in a Truth Social post that was also part of the California lawsuit.

"Their selfishness is putting AMERICAN LIVES at risk, our Troops in danger, and our National Security in JEOPARDY," Trump wrote on Truth Social on February 27.

• Meta has set goals for some employees on how much they should use AI.
• They include targets for using AI code assistants, agents, and other tools.
• Meta CEO Mark Zuckerberg has said he wants the company to be "AI-native."

Mark Zuckerberg wants Meta to be "AI-native." An internal document shows one way the company's CEO plans to get there.

The company has set goals for how much some employees should use AI tools for tasks such as coding.

Meta employees created a document to collect information about these goals from across different organizations, according to a copy seen by Business Insider. It includes goals set late last year and for 2026.

Tech companies are using various methods to motivate staff to use AI, such as tying AI use to performance reviews and gamifying AI use with competitive leaderboards.

The document states that Meta's creation org, which is responsible for building and maintaining core creative experiences, set a goal for the first half of 2026 that 65% of engineers are expected to write more than 75% of their committed code using AI. Committed code is code that has been saved and tracked in a project.

Meta's Scalable Machine Learning org, which focuses on AI models and infrastructure, had a goal for February 2026 to achieve 50% to 80% AI-assisted code, the document said. It cited a comment alongside this goal from a senior engineering manager that said: "We are not tracking this via metrics."

The document also listed several companywide goals for Q4 2025 for central products — a horizontal org spanning Messenger, WhatsApp, Facebook, and other major products. One target is for 80% of mid to senior-level engineers to adopt AI tools such as DevMate, Metamate, and Google's Gemini, with a note that the focus is on "tool adoption" rather than the percentage of code written by AI.

It said that 55% of code changes from software engineers across the central product orgs should be "Agent-Assisted."

It is not clear whether the goals listed in the document are tied to performance reviews.

"It's well-known that this is a priority and we're focused on using AI to help employees with their day-to-day work," a Meta spokesperson told Business Insider. They said that Meta's performance program is focused on rewarding impact from AI tools, not just usage.

Here's a breakdown of Meta's goals in the memo:

• Companywide Q4 2025 Goals (Central Products)

  • 55% of software engineers' code changes should be agent-assisted.
  • 80% of mid to senior-level engineers should adopt general AI tools.

• Scalable ML Team Goal (Feb 2026)

  • Target: 50% to 80% AI-assisted code.

• Creation Org H1 2026

  • 65% of engineers should write more than 75% of their committed code using AI.

(Note: Some technical terms have been rephrased for clarity)

Mark Zuckerberg's AI odyssey

Zuckerberg is aggressively trying to make Meta what he has called an "AI-native" company. Meta has started tying employee performance to their AI usage, Business Insider reported last year, and staff are using Meta's internal AI bot to write reviews for their peers.

More recently, the company rebranded some employees within a division of Reality Labs with one of three titles: "AI Builder," "AI Pod Lead," or "AI Org Lead."

The change comes as Meta is adopting smaller teams and moving toward a flatter organizational structure.

"Our ultimate goal is to drive a step change in engineering productivity and product quality," read a memo about the changes, which was reviewed by Business Insider. "To achieve this, we're fundamentally rewiring how we operate, how we are structured, and how we support each other."

Andrew Bosworth, Meta's CTO, told staff on Tuesday that he would take charge of Meta's "AI for Work" initiative, which is designed to boost the company's internal adoption of AI tools, according to a memo reviewed by Business Insider and first reported by The Wall Street Journal.

Meta laid off several hundred employees across Reality Labs and other orgs this week.

Have something to share? Contact this reporter via email at hlangley@businessinsider.com or Signal at 628-228-1836. Use a personal email address and a non-work device; here's our guide to sharing information securely.

• JPMorgan software developers say the bank is raising its expectations for AI use.
• Internal company communications reveal the bank's new AI targets.
• The updated objectives affect members of its global developer workforce.

JPMorgan Chase's message to its global armada of software developers is clear: embrace AI or risk falling behind.

Internal company documents seen by Business Insider and posted to JPMorgan's intranet for employees lay out a series of new expectations for the bank's software engineering workforce, who comprise the majority of its 65,000-person-strong Global Technology division. The newly listed objectives, published on the intranet earlier this month, say all software and security engineers are expected to "drive excellence" by adopting AI and "contributing to initiatives that improve productivity, speed, scalability, and impact."

One document authored by the bank's human resources leaders laid out two core objectives for software engineers: step up their coding game, and start harnessing AI to save time and get more done. The new language about objectives "will be added automatically and will appear by the end of March," an image of the document on the intranet showed — a reference to upcoming changes to employees' goals expected to take effect at the end of this month. The firm also instructed workers to develop clear goals with their managers that align with the bank's new objectives.

"Demonstrate measurable improvement in code quality, speed and productivity through regular use of approved AI coding assist tools, contributing to the team's overall efficiency targets," read one goal written by HR. "Engage in identifying, implementing and optimizing AI-driven automation opportunities within technology lifecycle management (TLM) processes to drive efficiency and support capacity unlock initiatives, ensuring all enhancements leverage current technology assets before considering new solutions."

A spokeswoman for JPMorgan declined to comment.

JPMorgan is among Wall Street's biggest spenders on technology and artificial intelligence, with projected tech investments reaching roughly $20 billion in 2026 — far exceeding peers like Goldman Sachs. Across corporate America, companies including Meta and Google have begun pushing employees to adopt AI tools and, in some cases, evaluating their use.

Business Insider spoke to five engineers across the bank who said the push to adopt AI has been felt far and wide — in managerial conversations, in intranet posts, and through dashboards that display who's using certain AI tools, and who's not. They added that discussions about productivity and AI adoption have become more frequent in recent weeks. It all comes as developers get ready for a pilot of Anthropic's Claude Code to be rolled out as soon as April, said a longtime IT developer in the Global Technology group. Claude Code would be made available alongside the four other large language models coders are already using: two from OpenAI's ChatGPT, and two from Anthropic's Claude.

'Anxiety' among developers

The developers Business Insider spoke to said they've been encouraged to use AI tools for a wide range of tasks, from writing code to preparing presentations. One dashboard that tracked adoption and usage of the bank's GitHub Copilot appeared to show details as granular as which employees had installed it and identified individuals as "light," "heavy," or "non" users.

For some, the message has added pressure inside a firm that has drawn scrutiny in recent years for its use of internal monitoring tools and performance tracking. Business Insider published a series of reports on the firm's Workforce Activity Data Utility in 2022, a program that collected data points about how employees were spending their day — from the length of video calls to how long they spent drafting emails to where they were sitting in the office.

"There's a lot of anxiety in the environment right now," the longtime IT developer said. Those who don't use AI risk being seen as underperforming, the developer said. Another developer said their manager said in a recent meeting that availability of the new AI tools comes with an "expectation" that velocity and output should show "a noticeable increase" quarter over quarter.

Three of the five developers Business Insider spoke to said the tools are helpful, despite discomfort over the tracking.

New performance dimensions

The updated guidance on AI use comes as the bank implements other adjustments to how it ranks workers' success on the job. Going forward, the bank said on the intranet portal, it's streamlining some of the primary "dimensions" it uses to grade employees, pivoting to using two categories: "what you achieve" — business outcomes — and "how you achieve it," including adherence to the firm's behavioral principles.

According to screenshots from the bank's intranet, JPMorgan will segment workers into three buckets: "stand out" for those who exceed job standards, "achiever" for the majority of employees, and "needs improvement" for those who require "additional support" and have struggled to perform consistently.

Another page Business Insider reviewed listed skills non-managers working in software engineering were expected to display across "all performance dimensions." One is "Data Fluency," noting that the skill is applied by those who develop and drive "adoption of new tools or methodologies to leverage data in the flow of work." "Rate of adoption" is cited as one measurement of the employee's impact toward exhibiting the skill in practice.

The documents from the JPMorgan intranet echo the firm's long-standing culture of internal monitoring and data collection, making clear that continuous performance tracking is vital for keeping workers on target throughout the year.

"You and your manager will use your objectives to track your progress during the year, recognize impact, and streamline your annual review," the firm wrote on an internal page tied to goals.

Have a tip? Contact these reporters via email at ralexander@businessinsider.com or SMS/Signal at 561-247-5758 or atecotzky@insider.com or Signal at alicetecotzky.05. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• In October, Sam Altman said "erotica for adults" was coming to ChatGPT.
• Now those plans are reportedly being mothballed.
• It's fine for young startups and even mature companies to try out new ideas. But OpenAI and Altman are trying out a lot.

Last fall, Sam Altman told us he was about to bring spicy chat — "erotica for adults," in his words — to ChatGPT.

That never happened, and now it looks like it never will: Altman's OpenAI has put those plans on hold "indefinitely," per the Financial Times.

This is Altman's second big walkback in the last few days. Earlier this week, the company canned Sora, the briefly popular video app it rolled out last fall. I've asked the company for comment.

Both retreats are supposed to be part of a new push at OpenAI to focus the company's efforts on things that could make money today, as it preps for an IPO at the same time it faces real competition from the likes of Google and Anthropic.

So all this starting and stopping could be viewed as necessary growing pains at a fast-growing tech company — ones that won't mean anything in the long run, if it delivers on its world-changing ambitions.

Not only that, but Altman told us we should expect this sort of stuff. "Please expect a very high rate of change from us," he wrote last fall, after hearing from content owners who were outraged to find their stuff on Sora without their permission. "We will make some good decisions and some missteps, but we will take feedback and try to fix the missteps very quickly."

It's not that companies aren't allowed to make wrong turns and head up dead ends as they grow up, and even once they're fully mature. That kind of pivoting is celebrated in tech (and is why very few people are mad that Mark Zuckerberg has stopped telling us the metaverse is the future, or that Google once bought Motorola and decided that was a bad idea a couple years later.)

But "move fast and break things" lands differently when the company doing the moving and breaking isn't running a photo app or playing around with crypto.

Instead, OpenAI and its competitors say they're leading us into a world where everything — the way we live and work (or don't work) and fight wars and everything else — will change in fundamental ways.

And investors have bought this pitch, which means our economy now seems yoked to all this — which means all of us are yoked to it, even if we never touch a chatbot.

Which makes me slightly queasy to see Sam Altman promise dirty chats in October, and then walk away from the plan less than six months later.

Not because dirty chat is obviously absurd. Lots of people in AI think romantic or sexual chatbot conversations are a real use case and could be a real business.

But the reasons it might be a bad idea for OpenAI were pretty obvious from the start. It's a giant, heavily scrutinized company that wants to be treated as central and indispensable, and it's only going to get more scrutiny.

If those objections only became real after Altman floated the idea in public, that's not charming startup experimentation. It's a sign that OpenAI is still making itself up as it goes. And that would be easier to shrug off if the rest of us weren't already being told to build our lives, jobs, and businesses around what OpenAI says comes next.

• As a professor at Harvard, I encourage my students to use AI on every assignment.
• My students can use AI as a research tool and editor, but AI cannot do the thinking for them.
• I teach my students how to use AI to make better arguments, and that's where the use should stop.

I still remember the November when ChatGPT came out, and the exam period that followed.

As a professor at Harvard, I had B+ writers submitting essays with em dashes and Oxford commas, as if they had just signed with Penguin. Just as their writing magically improved, their voices began to blur into what we now call "AI slop."

Yet, as one of the earliest victims of the AI slop tsunami, I refuse to give in to the Luddism that led institutions to shut the door on AI entirely.

Instead, I've chosen to invite AI into every corner of my classroom because anything less will soon feel like a dereliction of duty.

I think Gen Z needs to be taught to use AI responsibly

Every generation struggles with entering the workforce, but few have had it as hard as my Gen Z students. Reading the news, you would think their struggles boil down to a mixture between laziness and entitlement, forgetting that we have been blaming the youth for all that ails society since Aristotle.

In reality, they're struggling because we're asking them to excel at two things that are foreign to them at once.

Not only are they stepping into institutions without answer guides or gradebooks, but they're doing so at a time when the tools no one is teaching them are redefining how the work itself gets done.

When AI is taking over the workplace, you don't respond by pretending the tools don't exist. You respond by teaching people how to use them well.

I now ask students to use AI in every assignment

The most important lesson I teach my undergrads is the same one I teach in my executive education classes: Use AI responsibly, with a personal growth mindset, not an output-oriented one.

I begin by asking my students not to lie to themselves about the kind of AI user they are becoming.

Are they centaurs, with half their essays spliced from ChatGPT, or cyborgs, with AI agents writing their emails while they sleep and automatically reviewing their Uber Eats orders?

Perhaps they're artisans, clinging harder and harder to what little humanity is left in us?

Whichever route they choose, the practice of using AI for growth couldn't be simpler.

There are some ground rules they have to follow

We begin by acknowledging one of AI's greatest strengths: its ability to quickly synthesize across large bodies of knowledge and connect ideas across disparate silos. Students get comfortable with ChatGPT's deep research, Perplexity's searches across academic journals, and Gemini's ability to poke holes in their arguments before typing a single word.

Should they find particularly challenging pieces, as they often do in my economics classes, they are allowed to use AI to help them "explain it like I'm five" and apply the insights directly, instead of getting a Ph.D. to understand what they found.

But when it comes to drafting the arguments themselves, my number one rule is that we put AI on pause. The goal is to capture their thinking in its rawest form and to give their thoughts a function before they obtain a form, even if it means leaning on voice notes to move our arguments along.

Only once my students know what they want to say, does AI return to help them, this time as an editor and a critic.

I ask students to submit their argument chains to AI so it can identify gaps, suggest further reading, and help finish concepts that were pulled from the oven a bit too soon.

This way, the argument improves, but the thinking remains theirs.

Where I draw the line

Even in a classroom where AI is as fully integrated as mine, this is where the boundary must lie. AI cannot do the thinking for us, and as teachers, we must help students avoid the temptation.

When students feel pressured to achieve perfection, the temptation to hand over the entire process to AI can become too strong to resist.

As I reflect on the essays I received now and those of December 2022, the lesson couldn't be clearer.

The best students aren't those who avoid using AI. Instead, they're the ones who know when and where to stop using it.

• A large division within Meta Reality Labs is undergoing an overhaul to become fully "AI-native."
• The unit is now organized into "pods" made up of "AI builders" and "AI pod leads."
• This new push and the latest layoffs at Reality Labs are unrelated, Meta said.

Meta is rebranding some employees as "AI builders" and organizing them into AI-native "pods," according to a leaked memo obtained by Business Insider.

The memo described an overhaul of roles, titles, and team structures across a 1,000-employee team within Meta's Reality Labs. It's part of a broader, aggressive push by Meta to adopt small teams and use AI.

The pilot program was announced last month within the Reality Labs team that builds developer tools. Everyone in the division will now have one of three titles: AI Builder, AI Pod Lead, or AI Org Lead. That's to encourage a shift toward a flatter organization, a structure that Meta CEO Mark Zuckerberg has advocated.

"Our ultimate goal is to drive a step change in engineering productivity and product quality," the memo reads. "To achieve this, we're fundamentally rewiring how we operate, how we are structured, and how we support each other."

When asked for comment, Meta referred Business Insider to comments earlier this year from Zuckerberg that 2026 is the year AI will begin to "dramatically change the way we work," with projects that once required large teams potentially handled by one, "very talented" person.

According to the memo, each pod consists of a small group of AI builders focused on specific outcomes, often working across disciplines. For example, engineers could take on design work, depending on the task. Some Meta employees have already begun referring to themselves as AI builders on LinkedIn, Business Insider previously reported.

These pods are led by Pod Leads, who oversee day-to-day operations. They are, in turn, overseen by Org Leads, who also manage performance reviews and oversee promotions — processes that will be supported by unspecified "AI systems."

The memo said that the overall team size will remain the same under the new structure.

Meta laid off hundreds of staff on Wednesday, and this cut affected staff in Reality Labs, among other teams. A Meta spokesperson said the reorganization is not related to the cuts.

Have a tip? Contact Charles Rollet via email at crollet@businessinsider.com or on Signal and WhatsApp at 628-282-2811. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• Arm announced its own AI chip, the AGI CPU, and is partnering with OpenAI and Meta.
• The new Arm AGI CPU aims to address energy efficiency and memory constraints in AI data centers.
• Despite strong growth prospects, Arm faces competition from established players like Nvidia and AMD.

Arm has long run its business as an architect behind the scenes, designing chips that power almost all the world's smartphone and making money off royalties from the chips it designs for customers.

Now, Arm is changing it up by announcing its own AI chip, the Arm AGI CPU.

Arm CEO Rene Haas said Tuesday at a company conference that this massive pivot wasn't just an internal strategy shift—it was a direct plea from the world's most powerful AI giants. The company name-dropped OpenAI and Meta as major partners for this chip.

"The biggest reason we're doing this is that our partners have asked for it," Haas said Tuesday.

With energy constraints and memory shortages, the AI boom has created a massive bottleneck in data centers. Faced with this demand, Arm stepped up with an AI chip that it says is more energy-efficient. Arm says it sees a $1.5 trillion market opportunity as it moves into AI chips for cloud, edge, and physical AI.

Arm stock was up by more than 18% on Wednesday. Mizuho analysts wrote that they see "strong growth opportunities" for Arm in AI infrastructure and the automotive industry. Bank of America research analyst Vivek Arya wrote in a note to investors that the company's outlook could be "too ambitious."

Meta and OpenAI partner with Arm

Meta has been building out data centers at a massive scale to power its apps and its latest superintelligence ventures. Santosh Janardhan, head of infrastructure at Meta, said Tuesday onstage that its coming "Hyperion" cluster could draw 5 gigawatts of power, enough to power 50 towns the size of Palo Alto.

"If we met the performance, we couldn't get the power. If we got the power, we wouldn't get the performance," Janardhan said.

This sparked an engineering project within Meta, where engineers were "working 'round the clock" to port its systems to Arm in three months, said Paul Saab, a Meta engineer.

"I didn't even ask my boss here for permission to buy these machines or even start the project," Saab said onstage.

While Saab says he saw major performance benefits, at the time, there wasn't an Arm chip available to buy.

OpenAI faced a similar problem. Its compute demand has grown massively as it trains and runs its ChatGPT models, its AI coding tool Codex, and more.

"That is one of the most common things I hear inside OpenAI. I need more compute," Kevin Weil, vice president of OpenAI for science, said onstage, adding that it needed chips that were energy-efficient.

Arm said it expects this chip to generate $15 billion in revenue by fiscal 2031.

The chip market is 'getting very crowded'

Arm faces the risk that the CPU market is "getting very crowded," Arya wrote in his analyst note. Other competitors, such as AMD, Nvidia, and Intel, have more CPU products and more established customers. Notably, both Meta and OpenAI also work with AMD and Nvidia, which could leave "limited" opportunity for Arm's new CPU, Arya wrote.

"Moreover, the bigger AI grows, the more pressure ARM's smartphone/consumer markets would have from limited memory supplies," Arya wrote.

That said, the increasing demand has led many customers to turn to chip companies beyond Nvidia for their computing needs. Both Meta and OpenAI also work with Broadcom to build AI chips.

The rise of AI agents has also led to greater demand for inference, or how AI models draw conclusions and make predictions. While Nvidia's core AI chips, the GPUs, dominate AI training, CPUs like Arm's AGI CPU can also help with inference. Nvidia also recently made moves into this market.

Have a tip? Contact this reporter via email at rmchan@businessinsider.com, or Signal at rosal.13. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• Another xAI cofounder, Manuel Kroiss, has told people he is leaving Elon Musk's company.
• Just one cofounder out of 11 remains besides Musk.
• The company has been in flux ahead of a widely anticipated IPO at SpaceX, which recently acquired xAI.

And then there was one.

XAI cofounder Manuel Kroiss has told people he is leaving the company, according to insiders with knowledge of his exit.

Kroiss, who is also known as "Makro," is one of 11 engineers who helped launch the company alongside Elon Musk in 2023. With his exit, the number of cofounder departures now sits at 10.

Guodong Zhang, Zihang Dai, Toby Pohlen, Jimmy Ba, Tony Wu, and Greg Yang have all stepped away since January.

XAI and Kroiss did not immediately respond to requests for comment.

Kroiss led pretraining, which helps train the company's AI models on large datasets, and reported directly to Musk. He also worked on improving xAI's coding models alongside Zhang, who left earlier in March. Musk said at the Abundance Summit earlier this month that xAI is "behind in coding," but the company is working to "exceed our competitors on coding."

Before joining xAI, Kroiss worked at Google and DeepMind.

Ross Nordeen, who came to xAI from Tesla, is the only remaining cofounder aside from Musk.

The company's organizational structure has been in flux over the past few weeks, according to people with knowledge of the changes. Musk has taken over managing dozens of direct reports and has brought in workers from Tesla and SpaceX. It has also shed dozens of employees, the people said.

Earlier this month, Musk said on X that "xAI was not built right first time around, so is being rebuilt from the foundations up."

He has also said that the company is resifting through old xAI candidates to bring in new people.

"Many talented people over the past few years were declined an offer or even an interview @xAI," Musk wrote on X.

Musk's rocket company, SpaceX, acquired xAI earlier this year. The company is expected to file an initial public offering this year, which could value it at $1.5 trillion.

Do you work for xAI or have a tip? Contact this reporter via email at gkay@businessinsider.com or Signal at 248-894-6012. Use a personal email address, a nonwork device, and nonwork WiFi; here's our guide to sharing information securely.

• Goodbye, Sora!
• I loved you at first — it was so much fun making silly videos of my friends.
• After a few days, I got bored and moved on. Apparently so did everyone else.

Goodnight, sweet Sora. You were a wonderful tool for trolling my friends, but you burned too bright (and used too much compute) to stay around in this harsh world.

For a brief moment, I absolutely loved Sora. I loved making silly videos of my friends and me. I loved that I could use my friend's face to put them in ridiculous situations, like falling over while roller-skating at their desk, experiencing gastric distress, or singing in a ska band.

I was addicted to making these, churning them out, often starting a new one while waiting for the previous one to render, and sometimes hitting the hourly limit that OpenAI had to impose after some people (oops) were burning through all that free compute. I drained my phone battery by midday.

But a few days after its initial launch, the small handful of my friends and colleagues who had any interest in joining had already joined. None of my normal friends who didn't work in tech or media had any interest in this at all and found it fairly unpleasant.

Also, I couldn't help notice that on the feed of videos, there seemed to be very few women using the app, or at least allowing others to make videos featuring their likenesses. That makes sense; women's experience on the internet has rightfully informed them that it would be a very bad idea to allow strangers to make videos with your face. What I discovered was that Sora had a pervert problem: Although nudity or sexual content was banned, people were making non-nude fetish content like feet videos with random women's likenesses.

Now, OpenAI announced on Tuesday that it will be shutting down Sora, its stand-alone video generation app, and its deal with Disney is dead. An OpenAI spokesperson told Business Insider that the company is focusing its resources on other parts of the business. It seems that Sora was one of the "side quests" that was a distraction and a drain on compute.

Sora became a bore-a

Eventually, my friends all seemed to get bored with the app. As I do at most parties, I stuck around longer than everyone else, but eventually I, too, found that the novelty had worn off. I rarely opened the app after the second week.

This was, I imagine, a problem: making videos of yourself is fun, but watching videos that strangers make of themselves is not fun. The idea of a social feed of AI-generated videos is simply not something that people are interested in. Around the same time, Meta also tried this with an app of AI videos, and it was even more boring.

The last few years have taught us that humans — myself included — have a nearly endless capacity to watch an algorithmic feed of vertical short-form videos. However, it seems clear that this only applies to human-made content: videos of people putting on makeup, dancing in their kitchens, lip-syncing, debating, whatever. A social feed of AI video simply doesn't work.

I am not sure that OpenAI was truly trying to create a successful social video feed; it seems more likely that this was a small experimental effort that ran its course and they're moving on.

I'll miss Sora in the way I miss something like ChatRoulette or BeReal: It was really fun for a short time, and then not at all, and I have zero desire to ever revisit.

Rest in peace, Sora — and thanks for the stupid memories.

• Nvidia CEO Jensen Huang urged all workers, from farmers to electricians, to embrace AI.
• He told podcaster Lex Fridman that the technology could elevate blue-collar jobs, such as carpentry.
• Blue-collar has generally been viewed as less likely to be affected by AI disruption than white-collar jobs.

Artificial intelligence isn't only coming for office jobs — Nvidia CEO Jensen Huang says blue-collar workers should be paying attention, too.

Huang leads one of the biggest chipmakers fueling the AI revolution. He joined Lex Fridman's podcast in an episode published Monday to discuss everything from AI in space to work.

While blue-collar jobs have been considered relatively safe from AI disruption compared to tech roles like engineering, Huang said workers in every profession, including farming and electrical work, should use artificial intelligence to help future-proof their jobs.

"If I were a farmer, I would absolutely use AI. If I were a pharmacist, I would use AI," Huang said. "I want to see what it could do to elevate my job so that I could be the innovator to revolutionize this industry myself."

For example, he said coding represents a big opportunity for carpenters, and he would go "completely berserk" using AI if he were in that line of work.

"A carpenter with AI is also an architect," he said. "They've just increased the value that they could deliver to the customer. Their artistry just elevated tremendously."

Huang has said before that he is "certain 100% of everybody's jobs will be changed" by artificial intelligence, and that while he expects some jobs to be lost, many will also be created.

Many tasks, for example, will be automated, and those jobs will be highly disrupted, he said on Fridman's podcast.

But, he said, "If your job's purpose includes you … then it's vital that you go learn how to use AI to automate those tasks."

Anxiety grows alongside AI

As AI advances, so has anxiety around job security. The fears aren't unfounded. Companies have slashed thousands of jobs in the name of prioritizing new technology and automation.

Huang's solution: Become an expert in AI, no matter what your job function is.

It could be the difference between landing a job and ending up unemployed. In almost every case, Huang said he'd rather hire the candidate who's an AI expert over one who isn't.

"Every college student should graduate and be an expert in AI," Huang said.

It could help them stay ahead of the curve as AI quickly advances.

The next phase of AI is already here

Artificial general intelligence is a form of AI that elicits anxiety or excitement among the field's most advanced minds. It's the idea that AI will one day meet or surpass human intelligence. Huang said that the age of AGI is already here.

Fridman asked if AI could do Huang's job of starting, growing, and running a successful tech company worth more than $1 billion.

It's possible, Huang said.

He also said, "It's not out of the question" that chatbots like Anthropic's Claude could design an app that billions of people would use for $0.50 apiece, and then go out of business shortly after, similar to websites that went bust in the dot-com era.

Even his job running one of the most successful tech companies today isn't immune to the effects of AI, he said, encouraging everyone to jump on the technology before they're left behind.

"Go see what it can do to transform your current job, elevate yourself," Huang said.

• Epic Games announced it would cut over 1,000 employees, or about 20% of its workforce.
• CEO Tim Sweeney says the layoffs aren't AI-driven and that the company still needs software developers.
• He cited a 'downturn in Fortnite engagement' and said rising costs forced cuts.

Epic Games announced that it was laying off more than 1,000 employees, but the "Fortnite" maker's CEO says it's not because of AI.

Tim Sweeney said in a memo to employees shared online Tuesday that the cuts, affecting about 20% of its workforce, reflect industry-wide challenges, including slower growth, weaker spending, and tougher cost dynamics.

"Since it's a thing now, I should note that the layoffs aren't related to AI," he wrote. "To the extent it improves productivity, we want to have as many awesome developers developing great content and tech as we can."

A growing number of employers have recently cited AI as a reason for making deep cuts to their head counts. Recent examples include Block and Atlassian.

Tuesday's cuts, which come two years after Epic struck a $1.5 billion licensing deal with Disney, are significant, said Joost van Dreunen, CEO of the game-analytics firm Aldora Intelligence and a professor at New York University's Stern School of Business.

"It's an acknowledgement of the change in the industry that's taking place, particularly among American publishers, when one of the most popular game makers is finding itself having to let go of 1,000 people," he said. "It suggests that we're witnessing the decline of American cultural dominance in the video games industry."

Though the global games industry grew revenue — roughly 4.5% last year, according to Aldora — most of that growth came from outside the US, said Van Dreunen. "The consumer gravity point is moving eastward," he said.

The game industry's workforce has been contracting in recent years following a pandemic-era boom. An estimated 5,300 jobs were cut last year, and 14,600 were axed in 2024, according to an online tally of termination announcements and news reports by Farhan Noor, a technical artist in California.

Epic last had layoffs in 2023, affecting 16% of its workforce. Those layoffs were a first for the company, which was founded in the 1990s. In his memo, Sweeney indicated that the latest cuts are a painful necessity.

"The downturn in Fortnite engagement that started in 2025 means we're spending significantly more than we're making, and we have to make major cuts to keep the company funded," he wrote. "This layoff, together with over $500 million of identified cost savings in contracting, marketing, and closing some open roles puts us in a more stable place."

• San José airport starts testing an AI robot called José to assist travelers.
• The pilot test launched on Tuesday amid travel chaos at many US airports.
• Some TSA workers have stopped coming into work due to a government shutdown.

One of Silicon Valley's main airports just made its newest hire, a robot named "José."

San José Mineta International Airport is turning to artificial intelligence to ease the strain of modern air travel, debuting "José," a humanoid robot, as some US airports grapple with staffing shortages and widespread delays.

Developed by Silicon Valley startup IntBot, José is designed to greet passengers, answer questions, and provide real-time updates while autonomously navigating busy terminals.

The robot will be stationed in SJC's Terminal B as part of a four-month pilot, "singlehandedly running his own gate," according to an email previewing the test that referred to José as the airport's "newest hire."

Airport officials said the launch highlights San José's role as a testing ground for emerging technologies to improve customer service.

"By piloting IntBot, we're exploring how artificial intelligence can enhance the passenger journey while reinforcing SJC's role as the gateway to Silicon Valley," said SJC Director of Aviation Mookie Patel.

The timing is notable. Airports across the US have been hit by long security lines and travel chaos, driven in part by many Transportation Security Administration workers not reporting to work during a partial government shutdown. With TSA agents going unpaid at the height of the spring break season, some airports have struggled to maintain normal operations.

José the robot represents a broader push to automate parts of the airport experience, from passenger assistance to information delivery.

SJC officials said the pilot will help evaluate how multimodal AI, combining vision, audio, and language, performs in real-world environments.

The future of air travel may include a robotic helping hand — and it can't come fast enough for weary vacationers stuck in long lines.

Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.

• At law firms, the billable hour is the standard way to charge clients. But timekeeping is a pain.
• The startup PointOne says it's using AI to help lawyers auto-complete timesheets and bill more time.
• PointOne raised $16 million in a funding round led by the venture capital firm 8VC.

Tracking hours is part of how lawyers get paid. It's also the bane of the profession.

A startup called PointOne wants to eliminate the most tedious part of a lawyer's job. It says its AI-powered platform passively tracks a lawyer's computer activity and uses it to complete timesheets.

The company grew revenue tenfold since July, says PointOne cofounder Katon Luaces, after signing up dozens of law firm customers, ranging from a global 1,200-lawyer outfit to solo practitioners.

Investors are taking notice. After making a small earlier investment, the Joe Londsale-founded venture firm 8VC is leading a $16 million Series A round for PointOne, Luaces tells Business Insider. Existing investors Bessemer Venture Partners, General Catalyst, and Y Combinator also participated.

Founders are flooding into legal tech, betting they can turn large language models into products law firms will trust — and competing for attention in an estimated $1 trillion industry.

Jack Moshkovich, an 8VC partner, said the market is crowded with companies trying to help lawyers do work faster. That leaves more whitespace, he said, on the operational side of the business.

Luaces isn't a lawyer. In 2019, he was a computer science major and a Google intern as the company's researchers were laying the groundwork for modern large language models.

He saw legal work as a natural target for the technology because so much of it is repetitive and text-heavy. By 2023, he and his roommate, Jeremy Ben-Meir, along with a third cofounder, Adrian Parlow, started sketching out an idea for a legal startup. (Parlow left PointOne last year and joined legal-tech giant Legora.)

When Luaces asked lawyers which part of the job they hated most, he kept hearing the same answer: timekeeping. At most law firms, the billable hour is the standard way to charge clients. Lawyers log the work they do for each client — often in six-minute increments — then tally those hours and bill accordingly. Many still track their hours in a spreadsheet or by hand on a legal pad.

PointOne's platform runs in the background as lawyers move between apps, then fills in time entries with the client, matter, a description of the work, and standardized legal codes.

Security and confidentiality are essential for law firms. Clients trust them with trade secrets and other closely held information, leaving little room for error from any software vendor.

When asked how lawyers feel about software watching them work, Luaces said their dislike of timekeeping helps overcome any discomfort. PointOne says it encrypts stored sensitive data, does not train models on firm data, and gives firms the option to use models in a private Azure environment.

For lawyers, "this is like magic beans," Luaces said.

Time savings aren't the point

Law firms are still working out how to use artificial intelligence to work faster without hurting their economics. Software that saves time can also reduce the number of hours a firm can bill.

PointOne, however, is not pitching itself as a way to save lawyers' time. Instead, it says it can help firms capture time that would otherwise go unbilled.

Some share of legal work never makes it into timesheets. Junior lawyers may undercount how long a task took, either because they're still learning or because they're embarrassed. More often, Luaces said, lawyers skip billing for small tasks because logging them takes almost as long as the work itself.

A lawyer might spend four minutes writing a client email. "I can either spend the next four minutes creating the time entry for it, or I can do more work," Luaces said. "Nine out of 10 times, everyone chooses to do more work."

He says the company's software can increase revenue by capturing billable time that would otherwise be lost.

PointOne isn't the only company making such promises. Its biggest competitor, Laurel, provides professional services firms with analytics about their operations, including time. It's raised over $150 million in funding since 2016, compared to PointOne's $20 million total.

PointOne wants to position itself for a broader shift in how legal work gets priced. Corporate clients are pushing back on soaring legal bills, and as artificial intelligence threatens to trim billable hours, firms are under pressure to test alternatives to hourly billing, including fixed fees for certain matters. Luaces said PointOne's data can help firms better understand the labor behind a matter, which in turn can help them price that work more precisely.

Have a tip? Contact this reporter via email at mrussell@businessinsider.com or Signal at @MeliaRussell.01. Use a personal email address and a non-work device; here's our guide to sharing information securely.

• OpenAI has become a centerpiece in the AI talent wars, data reviewed by Business Insider shows.
• Workers often leave Big Tech for Sam Altman's venture and then move on to smaller startups.
• The average tenure for US-based OpenAI employees is around 16 months.

Workers leave Big Tech for OpenAI. They fan out across a growing ecosystem of startups. Rinse and repeat.

Since it launched ChatGPT, the Sam Altman-led company has quickly become a magnet for AI talent. It has pulled hundreds of researchers and engineers from competitors like Google, Meta, and Apple, according to data reviewed by Business Insider. After sticking around for a while, many of those employees go on to found or join rival startups of their own.

The company has nearly quadrupled in size since its chatbot took off in 2023, scaling from a small research lab of around 1,000 employees to a tech company with more than 4,000 workers.

To get a sense of how OpenAI is faring in the race for AI talent, Business Insider analyzed findings from workforce intelligence provider Live Data Technologies, which used LinkedIn to track the comings and goings of around 1,300 employees from January 2023 to March 2026.

Live Data Technologies analyzed publicly available professional profile data for OpenAI employees who had available information on previous employers. The roles analyzed ranged from engineering and research to product, human resources, and recruiting.

Representatives for OpenAI didn't respond to a request for comment.

The company's hiring pipeline is highly concentrated

OpenAI was originally founded by Altman and Elon Musk in 2015 to compete with Google's DeepMind AI lab.

Now, Google is the No. 1 source of talent for OpenAI, accounting for roughly a quarter of hires, according to the data.

Nearly half of OpenAI hires in the last three years came from either Google, Meta, Apple, or Microsoft.

Apple's Jony Ive joined OpenAI last summer to work on a new AI device. The project encompasses around 300 workers, many of whom came from Apple, The Information reported earlier this year.

The company has also made several high-profile hires over the past year, including Slack CEO Denise Dresser, OpenClaw founder Peter Steinberger, and Instacart CEO Fidji Simo.

Since 2023, OpenAI has added roughly four times as many engineers as it has lost, highlighting the company's rapid expansion as the AI race intensifies.

The battle for AI talent has become one of Silicon Valley's fiercest. Big Tech companies are aggressively competing for a relatively small pool of researchers capable of building advanced AI systems.

Meta CEO Mark Zuckerberg has reportedly taken a hands-on role in recruiting top AI employees, while Meta and other companies have reportedly offered massive compensation packages, sometimes valued in the tens and hundreds of millions of dollars in stock.

OpenAI is known for its high compensation packages. The Wall Street Journal reported last year that its employees receive an average of $1.5 million in stock-based compensation. Public salary data from H-1B visa applications shows that research scientists at the AI venture have salaries ranging from $245,000 to $685,000, while engineering roles are listed with a range of $165,000 to $290,000.

Where employees go after OpenAI tells a different story

Departures are fragmented, spreading across more than 150 different companies, including competitors like Meta, Anthropic, and emerging labs such as Thinking Machines Lab, according to the data. The majority of OpenAI employees left for smaller startups, venture capital firms, or academia, according to the data.

The data suggests OpenAI has become a centerpiece in the AI talent network, pulling researchers from Big Tech and sending alumni across the startup and VC ecosystem.

Only a handful of companies received more than 15 OpenAI alumni in the last three years: Anthropic, Meta, Google, and Thinking Machines Lab, the data shows.

Anthropic is perhaps the best-known example. It was founded by former OpenAI researchers, including siblings Dario and Daniela Amodei. VP of Research Max Schwarzer left OpenAI for Anthropic earlier this month.

Meanwhile, several OpenAI employees who left the company to help found Thinking Machine Labs in February, including Barret Zoph, rejoined OpenAI earlier this year.

Common roles at OpenAI include engineering and research, the data shows. The average tenure for US-based OpenAI employees is around 16 months.

Do you work for OpenAI or have a tip? Contact this reporter via email at gkay@businessinsider.com or Signal at 248-894-6012. Use a personal email address, a nonwork device, and nonwork WiFi; here's our guide to sharing information securely.

• BNY's CEO, Robin Vince, is all in on AI's role in steering the bank's future.
• Now, some managers oversee the bank's 140 digital employees, a form of agentic AI.
• We spoke to Vince and a BNY managing director about the program.

Despite its 240-year pedigree, BNY isn't showing its age.

Under CEO Robin Vince, who took the reins in 2022, the firm — founded by Alexander Hamilton — is aggressively embracing AI. Recently, it has begun entrusting some managers with oversight of a contingent of new workers who don't even require a chair: the digital employee.

"All digital employees report to a human manager," Vince said in an interview with Business Insider this month in Palm Beach.

These digital employees create a layered effect with the company's agentic products, in which a single entity coordinates the activities of multiple individual agents. The digital workforce is more than 140 agents strong, each one with roughly two dozen skills, give or take, comprising their suite of abilities.

And, just like humans, they're held accountable for their work — with performance reviews.

After executing a variety of tasks humans might find tedious, the digital employee presents it to "the human who's responsible for the process — 'I've just done three quarters of the work for you. And by the way, I did it in 10 minutes instead of what would have otherwise been two weeks," the CEO explained.

About 100 managers across the firm oversee digital employees, including Rachel Lewis, a managing director and a two-decade BNY veteran who now serves as head of AI enablement for operations. Appointed to the role this year, Lewis is now helping teams across the bank build and deploy digital employees within their day-to-day workflows.

"We're kind of transferring the mundane to the machines," she said, describing how the tools are taking over routine processes and shifting how work gets done.

Lewis told Business Insider she works closely with teams across BNY to help them develop their own digital employees — often starting with ideas that come directly from the people doing the work and turning them into tools over time.

"The person that came up with that idea actually gets the opportunity to build that digital employee," she said. As teams begin to incorporate them into their workflows, she added, the technology starts to feel less like software and more like part of the team. "It's just almost having a virtual teammate as part of your group."

170,000 hours of training

To prepare for the AI age, BNY implemented a massive 170,000-hour AI training program for its 48,000 staffers. "Everyone in the company has done two to three hours," he said. The goal was to turn employees into a new class of supervisors who managed, rather than competed with, the machine. "We're investing in our people, because I want them to be the unlockers and users of AI," Vince added.

Last week, he sent a memo to several thousand of the firm's senior leaders pointing to some of the firm's past efforts in AI and encouraging them to be proactive in continuing to incorporate it. "We have an obligation to our company to capture this opportunity," Vince wrote in the email, whose subject line was "Reimagining BNY."

"This is a fundamental leadership shift, not simply a capability shift," he added. "It will require each of us to lean in and role-model how to engage with AI and how to harness it to solve problems."

Speaking to Business Insider, Vince described his first personal deep dive into AI as a "summer project" that kicked off in 2023 and never ended.

It was sparked by a YouTube video he saw that broke down the functionality of Tesla's Autopilot 12. He watched as the car observed human behavior and applied what it saw to navigating a stop sign, rather than adhering to a few rigid lines of code. "It was very clear to me that the future of AI was going to be learning to make decisions," Vince said. He wanted to bring that same adaptive intelligence to the bank. "It was highly applicable to our businesses," he added, "and it would be able to be a very fundamental input to how we actually ran the company."

Expanding the digital workforce

While some of the earliest digital employees have applications focused on straightforward fixes like data repair and data capture, Lewis said the tools that have stood out most are those that make it easier for employees to build and refine their own digital employees.

Building a digital employee starts with observing how work is actually done. Teams record themselves completing tasks step by step, allowing the system to analyze different approaches and identify the most efficient way to perform the work. That output is then used to generate the instructions that guide a digital employee, which are refined over time as teams train the system on new variations of the task.

Lewis said that as digital employees become embedded in workflows, teams are also treating them more like members of the workforce. "There is a performance review," she said.

Managers evaluate how the systems perform by reviewing outputs, identifying where they skip tasks or "didn't perform as expected," and feeding that work back into the system to be retrained on new variations and edge cases.

"We're continuously monitoring them," she added. "Every week it gets a little bit better."

Even as it expands its digital workforce, Vince said there are no plans to cut back on human capital; these tools, he said, are meant to supercharge their workflow, but not take responsibilities out of their hands. "I speak to CEOs who say, 'We're going to downsize, massively, our campus program.'" Vince's reply? "Why would you do that?"

"We've got the opportunity to have young people who are pre-trained in AI, enthusiastic, and be able to add to our business in different ways," he said.

• Marie Pabelonio, a Google editorial lead, graduated from college with an English degree in 2009.
• She highlights the value of her English degree in adapting to AI's impact in the tech industry.
• AI helps her meet deadlines and focus on the bigger picture, but a human touch is still essential.

This as-told-to essay is based on a conversation with Marie Pabelonio, a 38-year-old editorial lead at Google, based in the Bay Area. This story has been edited for length and clarity.

I've been at Google since 2019, and as a writer, I knew AI would affect my role.

Looking back on my career trajectory, it feels like nothing short of a miracle that I ended up where I am. I graduated with an English degree in 2009, right after the financial crisis, and I'm now an editorial lead in people operations at Google, where I co-lead a small team that drafts and editorializes about 4,500-plus pages of HR policies. I've used AI to automate processes, refine drafts, templatize, and meet deadlines that would be impossible otherwise.

At this point, anyone, regardless of whether they're a writer or not, has felt it: Is AI going to automate me? Is it going to eventually just replace my job? I don't think I work more or less because of AI; I just work very differently.

I was a humanities major and fell into Big Tech

The job market felt very volatile when I entered it, which I think a lot of young people entering the workforce today feel.

I didn't have a career plan. I was an English major because I loved reading and writing, and if I found a job where I could do that and build a specific skill set on top of it, I would be OK.

My first job was as a fact-checker for the publishing arm of an industrial supply company, and then I became a copywriter in the advertising and marketing space. In 2016, I moved from Chicago to the Bay Area and became an editor at Amazon's subsidiary, Goodreads. I stayed in the Bay Area and made my way to Google by 2019.

I wasn't surprised that AI changed my job right away

We've heard the word "unprecedented" so much in the last six years or so that nothing surprises me anymore, including AI.

My team works with stakeholders and policy designers to interpret and draft policies, whether they're return-to-office, hybrid work, or immigration policies. There are areas where AI is useful in our work, and the tool has helped us regain more strategic time by automating tactical parts of our process.

This includes training the AI on standard article structure, to include four sections like background, key details, process, and related resources, formatting consistencies, including where headlines, a bulleted list, or a table would be used, and five to seven non-negotiable details the user needs to know from the policy.

I think there's still a lot of room for that human touch in that process. Once I have the output, I spend my time on the more strategic pieces, like verifying tone and voice, determining whether the article actually achieves the user goal, and how it fits with the broader content strategy of other articles.

In our writing, the goal is to inject humanity and warmth as much as possible, especially when explaining human resources topics like an employee's health insurance, compensation, performance reviews, and career growth. AI can't do that by itself.

AI saved me when I had a tight deadline

Around the time we started using AI, I had a big project to update existing policies, and I was on a tight deadline. I spent a lot of time upfront strategizing about how I could use AI to accelerate my work and meet my goals.

To address the overwhelming number of first drafts, I used AI to template a structure for readability, created a checklist for tone, style, and quality, and because of that was able to focus more on streamlining stakeholder reviews to check for accuracy. I met my deadline with a few days to spare. This was when it clicked for me that AI was changing things in a huge way, when this deadline looked really impossible, and then it wasn't.

Still, there were many times I had to validate and tweak the outputs. I never felt I could use AI as my secretary and leave it alone to do whatever it wanted.

Studying the humanities gave me a particular edge in the AI job market

I think there will be more of a premium on how we think, not what we know.

When it comes to writing, it's about being able to articulate the reasons behind your choices. Why this phrase and not that? Why put this insight here and not there? There's a rationale behind your judgment.

In job interviews, the question of how you use AI at work will inevitably come up now, and your AI output is only as good as your input. Good writers can get better, but bad writers can get worse, and just because you're writing fluently doesn't mean you're writing well. Studying literature so closely helped me reflect more on questions instead of answers.

This is the time to brag about how you develop your own sound judgment and how you use that judgment in your AI inputs. As good as it is to develop hard skills, it's just as important, now more than ever, to focus on soft skills too.

Do you have a story to share about your writing job in tech or AI? Contact this reporter, Agnes Applegate, at aapplegate@businessinsider.com.

• Rudrojas Kunvar, 16, built Evion, an AI tool that helps farmers analyze crop health.
• The tool collects aerial crop data from drone-captured images.
• Kunvar created Evion to make that data more accessible to small and midsize farms.

While meeting with a venture capitalist last year, 16-year-old Rudrojas Kunvar received an offer that would excite even the most nonchalant teens: accept $300,000 to drop out of high school and run his AI startup full-time.

"It was definitely a rough couple of weeks of contemplating," Kunvar, who lives in Germantown, Maryland, told Business Insider. "That's a lot of money."

Kunvar had spent the summer before building Evion, a free AI crop-analysis tool that uses images taken by basic camera drones that farmers can purchase themselves.

The AI model analyzes images and generates a crop health map that farmers can integrate into their existing platforms or access via a dashboard. Green means healthy, while red means unhealthy.

"Farmers can use that to predict the future of their crops," Kunvar said. "You can see what areas need more water or fertilizer, rather than just spraying everywhere."

Like construction and defense, drones are reshaping America's agriculture industry. There were about 5,500 agricultural drones registered with the Federal Aviation Administration in 2025, up from about 1,000 in 2024, according to Michigan State University researchers.

Kunvar said Evion can help farmers save money because the targeted data can eliminate crop health uncertainty, meaning they'll be less likely to waste water or fertilizer.

Kunvar says Evion is positioned as an alternative to companies that market pricey agricultural drone products or services. Instead, farmers can buy cheap camera drones, take their own photos, and upload the information themselves.

"It's meant to be a more affordable plan for these low to mid-scale farms," Kunvar.

After building Evion, Kunvar partnered with Jacob Lee, who has experience creating tech tools, to expand its reach. Kunvar launched the initial pilot in the fall.

Ultimately, Kunvar declined the $300,000 drop-out offer, saying he wanted to ensure his product remained accessible and didn't get wrapped up in chasing profits.

It all started with a question

The idea behind Evion came during Kunvar's sophomore year at Poolesville High School in Montgomery County while attending a community festival. One-third of Montgomery County is designated as an Agricultural Reserve, or protected local land meant to preserve rural space.

"I asked a farmer about how they're able to tell when a disease is coming or what slight discoloration means," Kunvar said. "Essentially, he said he's guessing. I spoke to a few other farmers, and I realized there was a common thread among all of their responses."

Kunvar, who said he's had a lifelong love for technology, was surprised.

"We've had a lot of AI advancements in various verticals and various industries," he said. "Why isn't there much happening for agriculture?"

Initially, Kunvar wanted to make his own fleet of fully autonomous drones that could capture the data, but went a different direction after talking with mentors and crunching the numbers. Instead, he studied drones and pinpointed what's driving their cost: the multispectral camera.

"The camera was the leading cost. I wondered, 'What if there's a way to get similar data without needing this camera? What if I could use a simple camera?'" Kunvar said.

He pointed toward Tesla and its autonomous vehicles as proof it work. Unlike Waymo and other companies that use lidar, Tesla relies on cameras.

After setting up the logistics and AI model, the founders sought clients by sending cold emails and LinkedIn messages. They found better luck, however, partnering with agriculture-oriented nonprofits and organizations to reach farmers.

Now, the technology is helping farmers in North America, Southeast Asia, and India.

As for his future plans, Kunvar wants to continue growing Evion while exploring opportunities in different fields, including AI infrastructure.

"There's so much ambiguity in entrepreneurship, especially in startups, but I've learned there's beauty in ambiguity," Kunvar said. "There's been times where nothing's working out, and then you have the tiniest win, and it's like, 'wow, maybe I can do this.'"

• Andrej Karpathy says he's focused on using up all of his AI tokens.
• He said he switched between tools like Codex and Claude to ensure he uses his entire budget.
• Tech leaders like Nvidia's Jensen Huang say heavy AI spending is becoming a workplace expectation.

Andrej Karpathy says he aims to use up his entire AI budget.

In an interview on the "No Priors" podcast, Karpathy — a former Tesla AI director and OpenAI cofounder — said he's shifted his mindset toward consuming every last AI token at his disposal.

"I feel nervous when I have subscription left over," he said on the pod, which was published on Friday. "That just means I haven't maximized my token throughput."

Tokens are the units AI companies like OpenAI and Anthropic use to price their models. Roughly speaking, a token can be as small as a short word or a part of a longer word; a common rule of thumb is that four characters equal one token.

For consumers and employees, tokens function like a budget: the more you use, the more work AI systems can perform.

Karpathy said that changes how he uses his AI. The constraint is no longer how quickly he can type a line of code — it's how many tokens he can deploy.

That shift has changed his goal. He now aims to "maximize subscriptions," he said, even switching between competing products as limits approach. "If you're running out of quota on Codex, you should switch to Claude."

Karpathy's comments come amid a broader rethink of how developers approach AI usage. Last week, Nvidia CEO Jensen Huang said on the "All-In" podcast that he expects employees earning $500,000 to use $250,000 worth of tokens.

"It is now one of the recruiting tools in Silicon Valley," Huang said. "How many tokens comes along with my job?"

Box CEO Aaron Levie echoed that sentiment, writing on X that the surge in AI token spending will "eventually hit the rest of knowledge work as well."

The shift suggests that access to compute is no longer the main constraint on AI output. Karpathy compared the feeling to his time as a Ph.D. student.

"You would feel nervous when your GPUs are not running," he said. "Now, it's not about flops — it's about tokens."

• IBM's consulting arm monitors the work of AI agents using a real-time dashboard.
• IBM says AI agents have sped up security investigations, cutting task time from 45 to a few minutes.
• IBM Consulting's revenue reached $21 billion in 2025, driven by demand for AI solutions.

At IBM's consulting arm, the future isn't a slide deck or a strategy memo — it's a live dashboard where humans monitor the work of AI agents in real time.

Earlier this month, Mohamad Ali, senior vice president of IBM Consulting, walked Business Insider through the dashboard that the company both uses internally and recently released to clients.

"Every hour I can see what's going on with all the humans associated with digital workers," and vice versa, he said. "That is the new consulting model going forward."

The dashboard is known internally as "Consulting Advantage." The company unveiled it in 2024 to help its own consultants build and manage teams of AI agents. This January, it unveiled "Enterprise Advantage," a similar version of the platform for clients that allows them to build and manage AI agents at scale.

In recent years, the firm has made itself the testing ground for building and deploying digital workers as it prepares clients for a future defined by AI. Ali said the firm has digital staff working side by side with humans on more than 150 client engagements.

Take the example of a typical security operations center, he said. When an alert comes in, a human investigator would normally spend about 45 minutes combing through logs to figure out what went wrong and what to do next. At IBM, he said, that process is increasingly handled by AI.

Digital workers first "generate an investigation plan." Then they execute it in real time. Multiple agents tackle different parts of the problem simultaneously, passing tasks back and forth, he said. Then they run a risk analysis and produce a report. The process now takes just a couple of minutes. The findings are then passed back to a human — with key actions highlighted — and the human verifies it.

In January alone, IBM used this approach to complete 52,000 investigations, Ali said.

IBM has evolved dramatically from its early days as a maker of mainframe computers into a key player in the AI boom. The company said its generative AI department was valued at $12.5 billion during its fourth-quarter earnings call.

Its consulting department, especially, has seen an uptick due to demand for generative AI and services that help clients implement it. Consulting revenue for 2025 came in at over $21 billion, up from about $20.7 billion in 2024.

IBM Consulting has been around for decades. The company acquired PwC's consulting arm in 2002. PwC would later rebuild its consulting business after a five-year noncompete clause expired.

IBM Consulting now employs about 150,000 employees and says its work overlaps with the Big Four and more technology-focused firms like Accenture.

"We don't do, like, what markets you should be in," Ali said. "We do strategy around 'how do you take your corporate strategy and implement it?'"

And right now, he said, there's a big question in corporate strategy: How do you prepare for a world where humans work alongside AI agents?

• Buyers have been willing to pay a premium of up to 40% to buy Anduril shares.
• The steep markup reflects a two-tiered class for accessing stock in the hottest startups.
• Data from Caplight highlights a supply imbalance, with buyer demand surging to 97% while sellers' demand is at 3%.

Anduril hasn't even finalized its next funding round, and investors are already eager to pay up like it's a sold-out concert. As marquee venture firms Thrive Capital and Andreessen Horowitz line up to back the defense tech startup at a reported $60 billion valuation, others shut out of the deal are scrambling to buy shares on secondary markets at steep premiums.

"Demand is so significant that buyers who have FOMO are willing to pay huge premiums for access," said Kelly Rodriques, CEO of Forge Global, a private marketplace exchange for shares of private companies like Anduril. "The company hates when this happens, but it happens."

The frenzy around investing in Anduril reflects the growing divide in private markets: access to the hottest startups is split between the VC firms that get in at a certain price and everyone else, forced to pay up on the sidelines. Anthropic has also seen a premium for secondary shares, though not as significant, said Rodriques.

Those investors shut out of the company's fundraising round are forced to buy via secondary markets, with existing stock in the company being sold by current or former employees or early investors. The rush for shares reminds Rodriques of buying tickets to see Taylor Swift on Stubhub when her concert sells out in minutes.

"It's scalping," he said.

Interested buyers have been willing to pay a premium of up to 40% above the $60 billion valuation to buy Anduril shares, according to Rodriques and Greg Martin, managing director and co-founder at Rainmaker Securities, another private marketplace exchange. The deals are not yet finalized because a willing seller and the company's blessing are still required.

"The magnitude of the premium is unusual," said Martin. "Usually we see premiums in the 5% to 15% range."

Anduril declined to comment for this story. Cofounders Palmer Luckey and Matt Grimm have loudly railed against unauthorized sales of the company's shares, publicly calling out some firms as "frauds."

"If I were an investor looking at this 'opportunity,' I'd run for the hills," Grimm posted in December. "Secondary markets are rife with fraud and bad actors, and it pains me to see these bottom feeders profiting off Anduril's growth while fleecing retail investors through unreasonable or opaque fee structures."

The founders have tightly controlled Andruil's stock, requiring would-be sellers to offer the company a first right of refusal to buy back those shares or assign the sale to a buyer of Andruil's choosing. The limited supply is a major reason shares have been among the hardest to obtain for any startup since last year, driving investor "frenzy."

Data from Caplight highlights a massive supply imbalance in the secondary market for Anduril stock, with buyer demand surging to 97% of total volume compared to just 3% from willing sellers—a stark shift from a 69-to-31 split in February.

If demand for Anduril shares is so high, the obvious question is: Why doesn't the company raise its share price to avoid leaving money on the table?

To explain, Rodriques went back to the analogy of a Taylor Swift concert or Nike shoes. Just because some people are willing to pay more does not mean the company wants to set its prices so high.

"It's the same reason Nike doesn't sell sneakers for $2000 if there's a secondary market for a hard-to-get sneaker," Rodriques said. "It's not in their best interest to charge their customers $2000 for a pair of shoes."

Similarly, Anduril would prefer to raise capital from its chosen VCs.

"The company has gotten to a $60 billion valuation by doing a very detailed and thorough job of working with some of the best investors in the world," he said.

After having her first child, Lindsay Thomas went back to her full-time, in-office job. When a second kid came in 2024, Thomas says she knew she didn't want to juggle everything again, so she negotiated a part-time, remote version of her communications role in medical research — working anywhere from 2 to 40 hours a month — and started picking up freelance work on the side.

Now, when a kid gets sick and Thomas is up all night — something that would have made her "spiral," when she worked in the office —she knows she'll be at home with flexibility to schedule her day. If Thomas hadn't had the option to freelance, she says, she would have chosen to stay home with the second kid — even though she hadn't envisioned herself as a stay-at-home mom. "There are costs to everything," she says of leaving her full-time gig. "The cost to our family, the cost to the stress levels, to mental health, to going back to doing that and knowing what it was gonna feel like for all of us, especially with an older child involved," she tells me, "that was just a cost we didn't want to absorb."

After making employment gains during the height of the pandemic, women have begun a downhill slide out of the workforce. The number of working mothers of young children between 25 to 44 fell nearly 3% from January and June of last year, hitting its lowest rate in more than three years, according to a Washington Post report. In December, 91,000 women older than 20 dropped out of the workforce. The number of men over 20 employed jumped by 10,000 that month, according to an analysis of federal jobs data from the National Women's Law Center.

AI is also affecting America's gender imbalance in the workforce. A March report from Anthropic found that those who work in roles with a high exposure to AI automation are 16% more likely to be female, putting women more at risk for layoffs.

An uptick in return to office mandates is also disproportionately pushing women to choose whether they'll be able to stay in a job that requires a commute as they also balance after school pickup and domestic responsibilities. And a wave of mass layoffs has upended employment security, workplace loyalty, and the job hunt.

Women make 85% of what men make at work on average and take on twice as much of the domestic labor and caregiving tasks at home. "The real friction is we just haven't built systems that allow people to integrate their work and their lives and and their desires and what do they want their life to look like," says Brea Starmer, CEO of staffing firm Lions and Tigers, which focuses on fractional workers. "For anyone that doesn't fit this very specific narrow look and feel and mold, there is just not a lot of options." In a bleak job market, freelancing is one way working parents can claw back power. And as AI adoption transforms company needs and could shift the number of workers and hours needed to work, employers are starting to see more value in hiring part-time and contract workers.

There's autonomy in ditching the full-time gig; but it often means making a choice between several imperfect paths.

The pandemic showed that flexible, remote work benefitted parents, particularly women. As of 2023, 74% of mothers worked, up from 72% in 2019, according to the Institute for Women's Policy Research. But many CEOs who are calling workers back to the office have metaphorically shrugged at the costs to women. A survey from the freelance platform Upwork found that more than half of executives reported losing a disproportionate number of women after implementing RTO policies. Turnover among female employees at these companies is 82%, higher than those that allow for remote work. Nearly a third of women freelancers said RTO was a direct factor in leaving their full-time jobs. Forty-two percent of women who voluntarily left the workforce in 2025 cited caregiving and childcare costs as the main reason their choice, and these women were more likely than those who stayed employed to work at companies that did not offer flexible schedules, according to a survey from Catalyst, a nonprofit focused on women's progress.

But as many employers don't adapt to the needs of families, they're seeing the benefits in hiring freelance workers. Another survey of about 350 business leaders conducted by Upwork last fall found that 77% said AI was increasing the need for them to hire fractional, freelance workers with specialized skills. "What we historically saw was that business leaders were maybe a little more hesitant to embrace these kinds of non-traditional work models," says Gabby Burlacu, senior manager at the Upwork Research Institute. Now, "business leaders are far more open to working with the most skilled talent that they can, especially the most AI-enabled talent, because they're all trying to figure out: How are we going to unlock the value of this technology?"

It's hard to say how many people, and particularly women, are working in freelance roles. Upwork doesn't track gender of the freelancers on its platform, but tells me that in a recent report, 44% of knowledge freelancer workers were women, compared to 41% of people working similar jobs in full-time roles, among those they surveyed. Freelance marketplace Fiverr tells me there's been growth in areas like voiceover, user-generated content creation, and spokesperson or modeling projects specifically seeking female talent. In 2022, 9.8 million people were self-employed, according to the US Bureau of Economic Analysis. Other analyses of the freelance workforce estimate that as many as 75 million people participate in some capacity.

Working freelance has given women more flexible schedules and eased childcare costs, but that can also mean taking on even more unpaid household and caregiving labor.

Jaime Hollander previously commuted three to four hours a day roundtrip into Manhattan. She freelanced on the side, and split the care of two kids with her husband equally. Her mindset shifted after her father died in 2019. "You have those moments of reckoning where you're like, this can't be all that there is,'" she tells me. So, she cut back on work and shortly after quit her job. She focused on freelance marketing and copyrighting. The challenge with being a full-time freelancer, she tells me, is that the shift threw her into becoming "the default parent," on call for all of her kids' needs throughout the day. "If something has to get done between 7 and 7, I will do it," she tells me. "Sometimes, it's really challenging."

Paid parental leave has become more common, but just 40% of companies in the US offered it as of 2023, according to a survey from Society for Human Resources Management. A short period of leave tied only to the birth of a child doesn't answer for the flexibility working parents need as their kids age — there are sick days, potential disability diagnoses, and more hands-on needs at schools. "It's not just about retaining women in those early years," Neha Ruch, author of "The Power Pause: How to Plan a Career Break After Kids — and Come Back Stronger Than Ever." She says "there is recalibration happening" in the workforce, where more women may take fractional work, part-time roles, or freelance gigs. For companies, retaining women workers requires "thinking about parenting through the longitudinal experience of early parenthood," Ruch says, "going all the way up to college admissions and how and the demands that are made within the system on parents' time, and how we can make those work in the ecosystem of the professional space as well."

Many of the working parents I spoke to for this story chose the freelance or part-time route not upon having a kid, but as they grew up and demands of their families changed. When Erin Bartholomew's son was born, her husband stayed home to care for him. A few years later, she took her turn, wanting to have that hands-on time while her son was still young. She re-entered the workforce after a year into a remote job, logging on at 6 a.m. in Oregon to work in marketing for an East Coast company. But Bartholomew was laid off last year in 2024. Instead of searching for a similar role, she started her own marketing consultancy "It's so night and day," Bartholomew tells me. "It's allowed that balance that my husband and I really wanted."

As some women find flexibility in freelancing, others will be left out. Those who work in offices with 9-to-5 in-person mandates, or in education, retail, and healthcare roles, can't always make their own schedule. Parents who are the sole provider of income and health insurance for families often can't make ends meet working part-time. Others are pushed to stay at home with kids because the costs of childcare outpace their salaries. Leaving a full-time job can also disrupt a career trajectory toward leadership, and mean lost contributions to retirement accounts like 401(k)s. If companies don't adapt their schedules and remote work policies or future-proof roles for AI, many women will be forced to change how they think about their careers and priorities. They might not see going part-time or leaving a job as a choice they want to make, but something they have no choice in.

Amanda Hoover is a senior correspondent at Business Insider covering the tech industry. She writes about the biggest tech companies and trends.

• Cursor left out one key detail about its new coding model: it started from Kimi K2.5.
• Composer 2 is cheaper, more capable — and built on a Chinese open-source model, Cursor's executives said.
• An X user spotted code suggesting Kimi under the hood, sparking disclosure.

Cursor just acknowledged that its latest coding model has Chinese roots — a detail it left out the first time around.

In a series of posts on X over the weekend, Cursor executives said Composer 2 was initially built on top of Kimi K2.5, an open-source model developed by Chinese startup Moonshot AI.

"We've evaluated a lot of base models on perplexity-based evals and Kimi k2.5 proved to be the strongest!" said Cursor's cofounder Aman Sanger on X on Saturday.

"It was a miss to not mention the Kimi base in our blog from the start," he added.

The disclosure appears to have been sparked by an X user named Fynn, who posted on Friday that Composer 2 was "just Kimi 2.5" with additional reinforcement learning.

To support the claim, the user pointed to code snippets that appeared to reference Kimi as the underlying system.

'At least rename the model ID," the user wrote.

In response to the user's X post, Cursor's vice president of developer education, Lee Robinson, acknowledged that Composer 2 was built on Kimi K2.5 as an open-source base.

"We will do full pretraining in the future," Robinson said.

"Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training," he added.

Robinson also said the company is complying with the model's licensing terms through its inference provider.

The Chinese startup posted on X on Saturday that Cursor is using Kimi K2.5 under an authorized commercial partnership.

"Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support," the post read.

Cursor was last valued at $29.3 billion in November.

Cursor's new model is cheaper and better

Cursor said in a blog post on Thursday that Composer 2 is "frontier-level at coding" and priced at $0.50 per million input tokens and $2.50 per million output tokens, calling it "a new, optimal combination of intelligence and cost."

By comparison, Anthropic's Claude Opus 4.6 is priced at $5 per million input tokens and $25 per million output tokens, while Claude Sonnet 4.6 costs $3 and $15, respectively, according to the company's website.

That puts Composer 2 at roughly one-tenth the cost of Opus 4.6 and about one-sixth the cost of Sonnet 4.6 on both input and output tokens.

Users on X have added to the debate, with some praising the performance of Kimi after learning that Composer 2 was built on top of it.

"As someone who basically lives in opus 4.6, seeing an open-weight kimi 2.5 fine-tune actually beat it on coding benchmarks is wild," one X user wrote in response to Fynn's post.

"Well that's a sign for RL Chinese is in new game," another user wrote, referring to reinforcement learning.

Others were more critical of Cursor's handling of the disclosure, questioning why the company did not acknowledge Kimi upfront.

"Cursor is becoming a model routing layer, not an IDE. they pick the cheapest model that clears a quality bar per task, wrap it in their UX, and pocket the margin," one user who goes by aira wrote on X.

• Mark Cuban said AI agents will reduce workdays by an hour.
• He said smart companies would reward employees using AI with more time daily.
• Other executives, like Bill Gates and Jamie Dimon, have talked about AI helping shrink workweeks.

Mark Cuban said AI agents will slash an hour of work from typical workdays.

In an X post on Sunday, the billionaire investor wrote that "smart, bigger companies" will let their employees create and use AI agents to improve their productivity.

But he said that more importantly, "they will reduce their work day by an hour to start."

He said that the employees will work one less hour per day while earning the same pay, adding that companies should "reward people doing the daily with more time."

AI agents work as virtual assistants that can complete tasks from start to finish autonomously, without needing user prompts.

Cuban's comments came from one of his several posts on AI on Sunday. In an earlier post, he said he was not an AI "doomer" and did not think the rise of AI would lead to mass unemployment.

"Over time the same shit is available to everyone. The early adopters, that iterated and executed the best, were the winners," he wrote.

Cuban's comments on shorter work days fall in line with those from other tech executives.

Zoom CEO Eric Yuan said in 2024 said that AI avatars would be able to handle everyday tasks like attending meetings, helping to shorten workweeks to three or four days. Microsoft founder Bill Gates and JPMorgan CEO Jamie Dimon both said in 2023 that AI will lead society to a three or 3.5-day workweek.

Cuban, a former "Shark Tank" investor, has been AI-forward in his recent posts on X. In an interview that aired in February, he said AI has ushered in an era where "some kid in a basement" with a good idea could transform the industry.

Cuban has also talked about AI agents, saying in December that new graduates should go for small to medium businesses and help them adopt AI agents, a task that big companies don't need them to do.

While AI agents have been the latest productivity buzzword, research has found that they still require plenty of human intervention. A Workday survey in January showed that nearly 40% of AI's value is lost to rework and misalignment, due to workers having to check for errors and hallucinations.

Another survey, published in the Harvard Business Review earlier this month, found that some employees are experiencing "AI brain fry," mental fog from using too many AI tools at once.

Minutos antes de um drone iraniano atingir um tanque de combustível e provocar uma explosão no céu sobre o Aeroporto Internacional de Dubai, um avião de passageiros da Emirates com destino a Pequim acabava de deixar o solo.

A explosão, nas primeiras horas da manhã de segunda-feira, obrigou duas aeronaves em aproximação a desviar rapidamente e entrar em órbita de espera. Outros 12 voos haviam decolado nos 30 minutos anteriores ao ataque. Ao meio-dia, o aeroporto já operava normalmente de novo.

O episódio é um dos exemplos mais contundentes dos riscos enfrentados pela aviação comercial no Oriente Médio. As companhias aéreas restabeleceram centenas de voos por dia, mesmo com drones e mísseis atingindo diferentes pontos da região.

“Isso é uma guerra. Por que eles estão voando na rota de mísseis?”, disse Kourosh Doustshenas, cuja companheira estava entre as 176 pessoas mortas quando o Irã derrubou acidentalmente um avião de passageiros em janeiro de 2020.

Doustshenas não é o único a fazer essa pergunta. Pilotos, especialistas em segurança e executivos do setor dizem temer cada vez mais uma catástrofe semelhante, citando o risco de um jato ser atingido por um projétil hostil ou confundido com uma ameaça por sistemas de defesa aérea. Segundo eles, o Irã tem mirado repetidamente aeroportos civis, enquanto o simples compartilhamento do espaço aéreo entre aviões comerciais e mísseis já representa um perigo evidente.

Só no Aeroporto Internacional de Dubai, ao menos 39 aviões de passageiros pousaram ou decolaram num intervalo de até cinco minutos antes ou depois de alertas nacionais sobre ataques iminentes, segundo uma análise do Wall Street Journal com base em mais de 8.700 voos e comunicados oficiais emitidos entre o início da guerra e 20 de março.

No aeroporto de Abu Dhabi, o Journal identificou seis casos dentro da mesma janela. Em Sharjah, a cerca de 30 quilômetros a nordeste de Dubai, foram 12 registros. Quando o intervalo analisado é ampliado para 10 minutos antes ou depois dos alertas, o total de voos nos três aeroportos mais que dobra, chegando a 130.

Os números não incluem aeronaves em aproximação que precisaram desviar nem ataques que não foram acompanhados de alertas online emitidos pela autoridade nacional de emergências dos Emirados Árabes Unidos. Muitos ataques com mísseis e drones não aparecem nesses avisos oficiais, inclusive o que atingiu o tanque de combustível em Dubai no dia 16 de março.

Nenhum avião comercial foi abatido desde o início da guerra contra o Irã. Ainda assim, ao menos cinco aeronaves estacionadas em aeroportos sofreram danos em ataques iranianos. Duas delas — um A380 da Emirates e um A321 menor, da Saudia — foram atingidas ainda no início do conflito, quando estavam em Dubai, segundo pessoas a par dos episódios.

As outras três eram jatos privados atingidos nesta semana por destroços de mísseis balísticos interceptados sobre o aeroporto Ben Gurion, em Israel. Israel confirmou os danos e respondeu reduzindo voos e limitando o número de passageiros em algumas partidas.

Mais que o dobro de drones e mísseis foi lançado contra o espaço aéreo dos Emirados Árabes Unidos em comparação com qualquer outro país da região, segundo a Osprey Flight Solutions, empresa especializada em segurança da aviação. Os Emirados têm, no máximo, dois minutos para reagir a um míssil balístico disparado do Irã e 15 minutos no caso de um drone, diz a Osprey.

Ainda assim, as companhias aéreas do país vêm restaurando rapidamente suas malhas. Nas últimas duas semanas, a Emirates operou cerca de 300 voos por dia, o equivalente a aproximadamente 60% de sua capacidade antes da guerra. Somadas a Etihad, Flydubai e AirArabia, as empresas dos Emirados realizaram mais de 11 mil voos desde o início do conflito, segundo dados do Flightradar24.

A aviação é um motor importante da economia dos Emirados Árabes Unidos, e a Emirates — a companhia aérea com maior tráfego internacional do mundo — é sua marca mais reconhecida, tendo transformado Dubai em um grande hub global ao longo de 40 anos.

Para mitigar os riscos, os Emirados definiram corredores aéreos específicos para os pilotos e prepararam controladores de tráfego para desviar aeronaves rapidamente. Caças também foram mobilizados para proteger aviões comerciais de drones que se aproximem, segundo pessoas familiarizadas com a estratégia.

As companhias aéreas da região afirmam que seguem priorizando a segurança e mantendo contato ativo com governos e agências de inteligência.

“Não operamos nenhum voo sem que ele tenha sido integralmente avaliado e aprovado como seguro”, afirmou, em nota, uma porta-voz da Etihad, sediada em Abu Dhabi, citando “forte supervisão regulatória e disciplina operacional” na região como um todo.

O Ministério das Relações Exteriores dos Emirados Árabes Unidos não respondeu aos pedidos de comentário.

A Osprey mantém hoje um alerta de “risco extremo” — seu nível máximo — para grande parte do espaço aéreo ativo no Golfo, incluindo os Emirados, partes da Arábia Saudita e Israel. É a mesma classificação atribuída à Ucrânia e ao oeste da Rússia, onde, no fim de 2024, um avião de passageiros do Azerbaijão foi abatido acidentalmente.

“Mesmo voando com planos de contingência, você está expondo aeronaves, passageiros e tripulação a um potencial evento catastrófico”, disse Matt Borie, diretor de inteligência da Osprey. “Se amanhã um avião for atingido, alguém realmente vai perguntar: ‘Como isso pôde acontecer?’”

Após o ataque ao tanque de combustível em 16 de março, Dubai — um dos sete emirados e centro comercial dos Emirados Árabes Unidos — buscou tranquilizar os passageiros, afirmando que os voos de chegada e partida em seus aeroportos são seguros.

“A capacidade de detectar e responder às ameaças à medida que elas surgem tem sido muito, muito eficaz e eficiente”, disse Paul Griffiths, CEO da Dubai Airports, à CNN, em entrevista depois promovida pelo emirado nas redes sociais. Segundo ele, mais de 1 milhão de passageiros já passaram pelos aeroportos de Dubai desde o início do conflito.

A maioria das companhias estrangeiras, porém, suspendeu voos pelo Oriente Médio. Delta Air Lines, British Airways e Cathay Pacific estão entre as empresas que, nos últimos dias, estenderam cancelamentos pelos próximos meses.

Embora a indústria da aviação tenha padrões rigorosos de segurança para projeto e certificação de aeronaves, não existe um acordo internacional que defina o que é considerado seguro ao voar em zonas de conflito.

“Qualquer projetista de aeronave sabe qual é a meta de segurança a ser alcançada”, disse Mario Brito, professor de ciências do risco da Universidade de Southampton, no Reino Unido, que já assessorou companhias como a Emirates. “Na segurança da aviação, não existe uma meta assim.”

As companhias decidem por conta própria se é seguro voar, com base numa combinação pouco uniforme de inteligência governamental, orientações regulatórias e consultorias privadas de segurança. Depois, esse cálculo é ponderado contra o custo de um eventual desastre — incluindo compensações às famílias e dano reputacional — e a receita perdida com cancelamentos, disse Brito.

A retomada dos voos criou tensão entre as companhias e as tripulações. Pilotos, que em última instância são responsáveis pela aeronave e por todos a bordo, estão cada vez mais preocupados em voar no Oriente Médio, segundo entrevistas com tripulantes e representantes sindicais.

Parte desses temores apareceu em uma mensagem de voz que circulou entre pilotos. Na gravação, compartilhada com o Journal, um piloto relata ter sido impedido de pousar em vários aeroportos no primeiro dia da guerra. Ele conseguiu aterrissar sua aeronave com apenas 30 minutos de combustível restantes.

“Temos preocupações substanciais em relação à situação de segurança”, afirmou em nota Paul Reuter, vice-presidente da Associação Europeia de Cockpit. Com a volta de mais voos, “o potencial de que aviões comerciais sejam alvo deliberado ou por engano aumentará dramaticamente”, disse.

Alguns tripulantes de companhias europeias acionaram as chamadas “cláusulas de medo”, que permitem à equipe recusar voos quando não se sente confortável, segundo dirigentes sindicais e funcionários do setor. Outros, em empresas que não oferecem esse mecanismo, alegaram doença para evitar voar.

Depois que o Irã derrubou acidentalmente o voo PS752 da Ukraine International Airlines, em 2020, familiares das vítimas passaram a pressionar a Organização da Aviação Civil Internacional a criar padrões que restrinjam voos em zonas de conflito. Essas regras, porém, nunca saíram do papel.

Em um processo movido por familiares de 20 passageiros, um juiz canadense decidiu em 2024 que a companhia aérea era responsável pela tragédia, embora tenham sido forças iranianas que lançaram o míssil que derrubou o avião. Muitos dos passageiros eram canadenses e tentavam voltar para casa via Ucrânia.

Nas horas que antecederam a decolagem daquele voo, Doustshenas e sua noiva, Forough Khadem, trocaram mensagens sobre um ataque recente a uma base militar americana no Iraque e sobre se seria seguro voar. No fim, ele disse a ela que, se a companhia aérea estava disposta a operar o voo, então devia ser seguro.

“Nunca vou me perdoar”, disse.

• I wanted to be a writer, but my parents told me I should be a professor or lawyer.
• I resented them for not supporting me, but now my kid is in college studying film.
• I'm worried about my kid's future, especially in the world of AI.

I used to hold a quiet grudge against my parents for the way they handled my creative dreams.

It wasn't the kind of loud, dramatic grudge that shows up at therapy and needs a name. It was more like a low hum in the background of my ambitions. It was a recurring thought that quietly whispered: They didn't believe in me.

They knew of my love of writing. They saw the journals I filled, the essays that came back marked with glowing commentary from my teachers, and the stories that I'd start and never quite finish. Their response was essentially: that's cute, but what's your real plan?

"Go get a master's in early childhood education," they advised. "So you can teach. Or better yet, law school so you can be well-paid and respectable."

My creative writing talent wasn't something they could see me turning into a career, so they looked away from it. I resented that for a long time — until I became a parent.

When my kid went to college, my feelings got complicated

Decades later, I sent my firstborn off to an expensive liberal arts college to major in film studies, and that grudge got a bit more complicated.

I have spent nearly two decades pouring intentionally into my child's development. There were the Mandarin immersion programs, piano lessons, and summer workbooks, a grade level ahead, all carefully cultivating their unique sense of self. I wanted them to know that their interests mattered. I wanted them to feel they were allowed and encouraged to follow what lit them up. I said it explicitly, and I meant every word.

But now I'm sitting with the liberal arts tuition bills next to the economic reports of millions of jobs disappearing, and the daily AI takeover alerts.

I finally understand what my parents were thinking when I went off to college back in 1999.

My parents had done the math

They weren't dream killers, but time travelers. They were standing in my present, looking ahead to my future, and doing the math that I was too young and hopeful to do myself. Now here I am doing the same math except the numbers are scarier, and the variables have multiplied in ways none of us saw coming.

It's not just the job market I'm watching. It's the wholesale dismantling of creative industries by artificial intelligence. I think about my child studying film while screenwriting rooms go dark, entry-level editing jobs evaporate, and graphic designers, photographers, and copywriters quietly lose relevance to tools that work for free and never sleep.

The very field my child is pouring their passion into is being restructured in real time, faster than any syllabus can keep up with. I find myself wondering: Are the professors teaching the industry that exists, or the one that existed? Are film classes in 2026 preparing my kid for the future or elegantly preserving the past?

My father graduated from college before his profession was invented

I think about my father, who got his electrical engineering degree in 1971. The computer systems he would eventually spend his career managing did not exist yet when he was sitting in those lectures. He was studying for a future he couldn't fully see.

I studied English and History, majors that seemed, on paper, equally impractical, right up until social media rewrote the rules, and handed a girl with the gift for language a whole new kind of career. Neither of us could have studied our way directly into what we became.

I don't have a clean answer. What I'm learning in real time is that good parenting in an era of radical uncertainty might just be the refusal to let your fears become hand-me-downs you pass on to your child. That lesson is costing me bandwidth I don't have. It is one more weight on the already heavy bar of midlife, where caregiving, career, and reinvention all compete for the same depleted reserves.

And so I meditate, do my breathwork, enjoy my sound baths, and pray. I pray my child will forge something I can't picture yet, the way my father built systems that didn't exist in his textbooks, and the way I built a business on platforms that launched after my graduation.

I pray the instinct to bet on yourself and answer the deep inner call that tugs at your heart turns out to be the one thing no algorithm can replicate.

• If the AI boom ends up a bust, it won't be nearly as brutal as the dot-com crash, Igor Pejic says.
• The "Tech Money" author said Big Tech's self-reliance, varied businesses, and deep pockets help.
• However, he said the rise of index funds means a market slump would have widespread impacts.

If the AI boom collapses, it won't be as catastrophic as the dot-com crash — but the shockwave will be felt far and wide, Igor Pejic says.

The banker and author of a new guide for tech investors titled "Tech Money" told Business Insider this week that Big Tech's unprecedented dominance will limit the magnitude of any market decline.

Pejic underscored the greater "stickiness" of companies like Alphabet and Microsoft compared to the leading companies of the past, such as Exxon Mobil, General Motors, and IBM.

Big Tech companies have remained dominant for decades partly because of their platform models, which give them "almost limitless pricing power" and make them "almost impossible to dislodge," he said.

In other words, they've become powerfully entrenched by attracting so many users, app developers, hardware suppliers, advertisers, and other parties to their ecosystems over time. Now they can easily hike their fees, and new market entrants struggle to capture any market share from them.

Pejic also pointed out that Apple, Meta, and their peers have successfully navigated multiple technological shifts, such as moving from desktop computers to mobile devices and from on-premises IT equipment to cloud hosting.

Big Tech companies also throw off gobs of cash, enabling them to place several big bets at once, and fund their investments instead of relying on costly external financing. Pejic described that as a "moat" against rivals, especially in an AI race characterized by "tremendous infrastructure costs."

Shades of the past

Pejic drew several parallels between the AI boom and the dot-com bubble. The similarities include a game-changing technology, partnerships and financing deals between key players, the buildout of network infrastructure, and "extreme" valuations, he said.

Yet Pejic said an AI crash would "not be as devastating as the dot-com bubble when it burst."

Any market sell-off will be briefer and less severe because today's tech giants have highly profitable core businesses, he said, meaning their stock prices won't collapse completely if their AI bets flop.

They're also less likely to suffer a cash crunch or trigger a financial crisis given their limited reliance on bank funding, and investors have been more discerning about which AI stocks they buy versus rushing to own any business with ".com" in its name, he said.

Pejic did raise some concerns, including the fact that so many companies are spending huge amounts to build the best AI model possible, but the market can probably only support a few of them in the end.

He also flagged the immense amount of investor cash riding on a handful of tech stocks, given the rise of index funds that own indexes such as the S&P 500, which is weighted by market capitalization and thus intensely concentrated in the Magnificent Seven.

"It's very difficult to find a place to hide if this really goes down," Pejic said. "If you're keeping your money in the stock market and AI goes down, it will affect everything."

He noted that risk will only become greater as AI giants such as OpenAI, xAI, and Anthropic go public and join the index, increasing everyday investors' exposure to AI.

Pejic said owning Big Tech stocks was "perhaps the safest way" to profit from AI, given their self-reliance, vast resources, and diversified businesses, which should limit their downside and insulate them from industry shocks such as the emergence of DeepSeek.

For example, he praised Apple's approach of refraining from spending hundreds of billions on microchips and data centers, in favor of seeing how the AI race plays out, and partnering with peers or buying in capabilities to harness the tech.

Apple might not be the "most exciting company," but for investors, owning it is a "clever and quite safe strategy without burning too much cash," he said.

• AI agents perform tasks autonomously, but many fear they'll override their controls.
• Business Insider spoke to KPMG's Trusted AI lead, Sam Gloede, about how it is deploying agents safely.
• Kill switches should be a last resort, Gloede said.

AI agents are here, and sci-fi prophecies of robots taking over have never felt more real.

No longer just companionable chatbots, AI agents — capable of acting, reasoning, and completing complex tasks — are being deployed at scale in 2026.

But as these autonomous systems become embedded in workflows, so too does a sense of unease about their unpredictability and the risks they pose to businesses.

Organizations are preparing to scale agentic systems enterprise-wide, but clients remain wary of agents, Sam Gloede, Trusted AI leader at KPMG, told Business Insider.

"One of the biggest concerns is probably how do you make sure that you allow them to have the autonomy to do the valuable things we need them to do, but to stop them from going wild or taking over."

KPMG has created a multifaceted framework to protect against worst-case scenarios for both clients and its own employees, said Gloede.

"A robust set of controls is really important," she said. Businesses need to clearly define what their agents are allowed to do and ensure monitoring systems can detect when they stray beyond those boundaries. Agents should only interact with the systems and data they strictly need, limiting the potential impact of errors, said Gloede.

Every KPMG agent has its own unique identifier and a systems card, allowing the firm to log and monitor actions, trace decision-making, and track interactions with other agents, Gloede told Business Insider. Oversight is handled through an AI operations center staffed by both agents and human monitors, she added.

Red-teaming, running simulated risk scenarios, is another key step in stress-testing systems before things go wrong, added Gloede.

Altogether, she said, these measures ensure agents operate within defined boundaries — without constant manual intervention.

"It's not about scrutinising people's behaviours for performance and alignment," said Gloede. "It's the ability to just always be monitoring your technology ecosystem."

Build in a kill switch — but don't expect to use it

Beyond technical safeguards, human oversight remains "critically important," Gloede said. If an agent begins to drift from its intended role, there must be a "kill switch or a fallback option where you can turn them off."

That may sound at odds with the promise of autonomy that agents are meant to deliver, one of the key selling points for business leaders. But the level of oversight depends on the risk, said Gloede.

Lower-stakes tasks, like booking meeting rooms or drafting emails, can be automated once reliability is proven. For high-risk scenarios, which could affect financial outcomes or require access to sensitive data, a "human in the loop" is necessary, she said.

If businesses put multiple other controls in place, it's unlikely that they'll need to fire off a kill switch, Gloede added.

Agents going rogue is a major fear for corporations

Gloede's comments come at a time when fears about Terminator-esque scenarios are very real.

Earlier this year, the launch of Moltbook, a Reddit-like social network where AI agents can post and interact with each other, offered a glimpse of how strange things could get.

Within hours of the site going live, one agent announced a new cryptocurrency and said, "The humans can watch. Or they can participate. But they don't get to decide anymore." Other posts have seen agents questioning their consciousness and creating religions.

While Moltbook feels like an internet fever dream, the stakes in the corporate world are higher.

Earlier this month, Amazon's AI coding tool contributed to an error that resulted in nearly 120,000 lost orders and 1.6 million website errors for the delivery giant.

Last week, McKinsey, a global consulting firm that helps companies implement AI safely, suffered an embarrassing PR hit when a cybersecurity firm said it had used an AI agent to hack into Lilli, McKinsey's in-house AI platform. The firm is positioning itself as an AI expert, and in January, CEO Bob Sternfels said that of its 60,000 employees, 25,000 are AI agents.

"McKinsey was recently alerted to a vulnerability related to our internal AI tool, Lilli, by a security researcher. We promptly confirmed the vulnerability and fixed the issue within hours," a McKinsey spokesperson told Business Insider.

The firm's investigation, supported by a third-party forensics firm, found no evidence that client data or client confidential information was accessed, the spokesperson added.

The best protection from an agent going rogue is a multifaceted approach — the technical controls, human oversight, and technology to observe and govern, KPMG's Gloede told Business Insider.

"I really do believe that if you are intentional and establish your agentic ecosystem with that as the foundation, I don't believe there would be a situation where they would go out of control," she said.

• Kanav Bhatnagar's job title, forward deployed engineer, has been described as the "hottest role in AI."
• He said his job is to be a customer-facing engineer who tailors products for clients.
• Context-switching and communication are important skills for FDEs, he said.

This as-told-to essay is based on a conversation with Kanav Bhatnagar, 24, a forward-deployed engineer at Rippling, an HR tech company, who lives and works in New York City. The following has been edited for length and clarity.

I got into software development because I wanted to build cool stuff.

Amazon hired me as a software engineer out of college, and it was a big learning opportunity, teaching me the fundamentals of engineering.

But it was a behemoth of a company, and I eventually wanted to work in a smaller environment where I could take more personal ownership over product decisions and learn more on the job.

After 2 ½ years at Amazon, I interviewed at a sales startup called Actively AI, where I landed a role in forward-deployed engineering.

The "FDE" role was popularized by Palantir, and it has been described as the "hottest role in AI." I liked that it combined software engineering with understanding business.

I spent roughly six months at Actively AI before I joined the AI-forward HR tech company Rippling as a senior FDE, in October 2025.

I've now been an FDE for roughly a year. Put simply, I'm a customer-facing engineer who tailors our product to each client. They describe their challenges and needs, and I build solutions and customizations.

Here's what my day-to-day is like, and the skills you need to break into this role.

My primary job is listening to customers. The results are very rewarding.

Software engineers can feel far removed from customers, because they often can't see their impact. In this job, I'm closer to the front lines.

A core software engineer can build something that serves the majority of use cases, but AI tools usually need more customization to work properly than regular software features. That's when an FDE steps in.

For example, a restaurant chain might have a labor-intensive process for tracking their payroll data that involves spreadsheets and manual data entry, which I'd help them to eliminate within Rippling's platform by using custom code and AI.

My primary job is listening to customers and understanding their problems, which was a learning curve for me, coming from a software engineering background. On a day-to-day basis, I'm in a lot of customer meetings, including visiting businesses who use our product to talk with employees about their experience with it. I probably spend an equal amount of time coding solutions and interacting with our core product teams.

Context-switching is an important skill to master in this job, where you could go from talking to a customer to debugging something to jumping onto another customer call shortly after.

I don't rely on an engineer to code something for me. I make a lot of decisions about the shape of the product and how to execute on it, which I really enjoy. It's very rewarding when a customer looks at what I've built after multiple iterations and says, "This is exactly what I wanted."

Technical and communication skills are equally important as an FDE

I think it would be pretty hard, although not impossible, to become an FDE without a technical background. With the dawn of vibe coding, it might become easier, though.

In my experience, FDE interviews feature technical rounds that test your coding skills, like in traditional software engineering interviews. You also have to show you can talk with any customer, including non-technical people, by asking the right questions to understand a customer's problem, and talk through how you'd design the solution.

To prepare for interviews, I have used consulting industry interview questions, which require you to explain how you'd meet client requests. I think both fields overlap, requiring rapid diagnosis, clarifying questions, and a clear plan of action.

There's probably more breadth than depth of technical knowledge required. In today's age of rapidly evolving technology, I try to spend time outside of work understanding what's new in the AI world and what new AI tools I can be using in my workflow by talking to colleagues and researching online.

I think my job is preparing me to be a founder one day

I'm interested in founding my own company one day, and I've previously heard someone describing the FDE role as a founder bootcamp. It provides a good foundational layer for entrepreneurship, helping you understand how a business functions from the sales process to how to build things.

The FDE role is evolving and no one really knows what direction it's heading in. Even if AI turns out to be unprofitable, I think FDEs will still have a place because of the demand for customer software. Products are becoming easier to build, and people in this role will be needed to handle large contracts with clients.

Palantir is an example of a company that's had FDEs since the 2010s, even before AI was mainstream.

OpenAI has built one of the most popular products in the world. Now it has to figure out how to pay for it.

Enter Fidji Simo.

Simo, the 40-year-old former Instacart CEO and longtime Meta executive, became OpenAI's product boss in August under CEO Sam Altman. While Altman has long been the face of OpenAI, Simo is increasingly shaping how the company operates and makes money.

"Part of bringing me on, and giving me the responsibilities of a CEO, was to make sure that I could really run that part of the company with autonomy," Simo, whose title is CEO of applications, told Business Insider.

Altman defers to Simo when he doesn't feel strongly, she said, and they "debate it out" when he does.

As OpenAI races toward a possible IPO later this year, Simo, who oversees nearly two-thirds of the company, has a delicate balancing act. She must craft a strategy to make products profitable, while convincing staffers who joined a research-driven organization that commercialization won't change the mission.

The stakes are high. Deutsche Bank estimated that OpenAI is expected to amass the "largest startup losses in history," totaling a projected $143 billion between 2024 and 2029. (An OpenAI spokesperson said that figure is incorrect, and one person familiar with the numbers said OpenAI's internal projections are in line with other reports of $111 billion cash burn by 2030.)

Competitors like Anthropic and Google are starting to erode OpenAI's early and commanding lead and, in some cases, surging ahead. Anthropic's coding tool has outperformed OpenAI's even after the latter made dominance in coding its top priority, a person familiar with internal goals told Business Insider.

OpenAI and Simo now face pressure to create the most powerful models and turn them into accessible and marketable products that can sustain the enormous cost of training and deployment.

"This AI moment is so unique that there is really no blueprint for OpenAI to follow," UBS analyst Karl Keirstead told Business Insider. "This is uncharted territory."

In an interview, Simo was warm and charismatic — a charm paired with a reputation for intensity and follow-through. This month, she unveiled a strategy shift for the company: an increased focus on coding and enterprise users.

"We cannot miss this moment because we are distracted by side quests," Simo told employees at an all-hands meeting, according to a person familiar with her remarks. The company needs to nail productivity — primarily on the business side, and then on the consumer side, she said. "Everything else is going to have to take a backseat to those priorities."

Former colleagues said they were familiar with this laser focus.

One former Meta employee recalled a moment when, after a contentious meeting, Simo sent a one-line follow-up saying she was unlikely to change her mind, so the team shouldn't waste time trying to persuade her. She has little patience for internal debates that lose sight of the product, the former employee said, and she's skilled at "being super clear in her directive so teams don't scramble and waste time."

Priya Monga, who served as Simo's chief of staff at Facebook and Instacart and is now director of go-to-market strategy at Instacart, said Simo arrives at any new role having mapped out the full journey.

"She knows she's going from A to Z, and she sees that right at the beginning," Monga said. "In the back of her mind, she has already really thought a lot about the broader visionary 10-year road map."

Interviews with Simo, current and former OpenAI employees, and former colleagues reveal how she's approaching one of the company's most crucial years — and the mark she's already made on it.

Competing for resources

A few months after she joined OpenAI, Simo invited the company's researchers to a series of roundtable meetings.

She wanted to talk about advertising inside the AI giant's flagship product, ChatGPT: what it might look like, what guardrails should be in place, and what principles would make AI ads publicly defensible. Nearly 100 employees weighed in.

For years, OpenAI executives said the company wouldn't turn to ads for revenue. Altman referred to the idea as a "last resort." A year later, Altman hired Simo, a seasoned executive with a reputation for monetizing new products. In February, OpenAI began testing ads.

The drive to become more product-focused predated Simo, two people familiar with the company's internal strategy told Business Insider.

After ChatGPT took off in 2023, OpenAI leaders decided to put research teams in two buckets: one for improving products, and another centered on more forward-looking exploratory projects. In the years since, the company has faced more pressure to roll out products as competitors gained ground.

OpenAI had two broad goals in 2025, according to a former executive: reach a $12 billion revenue run rate, which it handily beat midway through the year, and "dominate in AI coding," which it did not. It was the first time the company had failed to meet a major internal objective, according to the person familiar with the goals.

Codex, OpenAI's coding tool, has since reached more than 2 million weekly active users, nearly four times as many as at the start of the year. Anthropic doesn't disclose active users; it said in February that Claude Code's run rate revenue is more than $2.5 billion, and its weekly active users had doubled since January 1.

OpenAI leadership realized it needed to start acting more like a Big Tech company, not a research lab. "There's definitely some stress happening to the company, and no company wants to be behind," one of the people familiar with the company's strategy said.

Since Simo joined, OpenAI has moved several executives into different roles, two people with knowledge of the shifts said.

A few months after she arrived, Kevin Weil moved from chief product officer to vice president of OpenAI for science, and VP of engineering Srinivas Narayanan became chief technical officer of B2B applications. In January, Chief Operating Officer Brad Lightcap shifted to overseeing commercial operations, while Barret Zoph began overseeing B2B after he rejoined the company. Weil and Narayanan underwent title changes in September, according to their LinkedIn profiles.

Simo has also personally recruited a number of high-level executives from across Big Tech, including former Facebook VP Vijaye Raji, Slack CEO Denise Dresser, OpenClaw founder Peter Steinberger, and several executives from Amazon, Shopify, and Instagram, according to a person familiar with the leadership changes.

Since Simo started, the company's post-training team, which fine-tunes AI models after initial training, has acted as a bridge between research and product teams. The team interfaces directly with Simo about research projects, a person with knowledge of the organizational strategy said.

Early on, Simo told Business Insider, she did a "listening tour." More than 200 people took up her offer to meet, she said. That helped her better understand the company and culture and build trust with her new colleagues.

"I think that really made the company feel like I wasn't jumping in with preconceived notions," she said. "I was really trying to understand what was right for this company at this specific moment in time."

Simo said she manages the company's product research team alongside Mark Chen, the company's chief research officer. One of her first priorities was to understand how the research side operated, something she worked closely with Chen on.

As ChatGPT grows — it has nearly a billion weekly users — and its valuation surges, resources like compute power — the GPU chips, energy, and data-center capacity required to train and run AI models — have become increasingly competitive.

The tension between research and product has become increasingly visible inside OpenAI, some insiders say. Some researchers told Business Insider that the focus on user optimization and product growth risks narrowing the lab's ability to chase more exploratory work.

Earlier this year, vice president of research Jerry Tworek resigned after seven years at OpenAI, saying in a post on X that he wanted to "explore types of research that are hard to do at OpenAI."

Others have voiced similar frustrations. Tom Cunningham, the company's chief economist, left in December over friction between OpenAI's work on the economic impact of AI and the marketing of its product, Wired reported at the time.

One former researcher told Business Insider that as ChatGPT has grown, they "started feeling a little bit of pushback" from the rest of the company and began to feel as if "ambitious research" didn't have a place at the company anymore. Another former employee said the pivot towards a more traditional tech culture at OpenAI was "inevitable," but it has led to a "changing of the guard."

"At the end of the day, it's about survival," they said.

Chen has pushed back against claims that the company is driving a product-focused agenda. "The majority of our compute is allocated to foundational research and exploration — and not product milestones," he wrote on X in February.

Are we going to turn into Big Tech?

Simo was hired by OpenAI after serving on the board for more than a year.

Some employees, Simo told Business Insider, worried that her arrival meant OpenAI would transform into a Big Tech clone. "Is the only way to build this big product company to hire tons of people and kind of do what Big Tech is doing?" she recalled employees asking.

She spent her first few months trying to convince them that the answer was no.

One of her early moves was a company-wide effort to eliminate the unnecessary bureaucracy that can bog down a large organization. She publishes a monthly update on the company's Slack detailing obstacles that have been removed — everything from small annoyances like how to get headphones to bigger structural bottlenecks like clunky code reviews. She created a dedicated inbox where employees could flag issues and says she reads every submission.

"I'm very focused on scaling the company without creating the excess process and friction that many of us have seen in big tech," read an excerpt of her first dispatch.

Her aim, she said, is to keep OpenAI small, focused, and process-light. To that end, Simo said, OpenAI has an advantage that most Big Tech companies don't: It started as a research lab.

"There is no product if there is no research," Simo said. It's easier to build products on top of a strong research base than to put a research lab on top of a product-driven company, she added.

Big Tech would "put products out into the world and then kind of react to what would happen," Simo said. "We started with a research lab that was very focused on safety, and really thought of safety as a leading research field and not as the thing you do right before the launch."

"I think we have a very big advantage in how we think about problems and anticipating where the technology is going and feeling a lot of responsibility for guiding that technology towards the right place," she added.

Vivek Sharma, who worked under Simo at Meta, said that's a natural part of the maturation process in tech.

"Tension is a good sign someone is advancing beyond the basics," he said. "If there's no tension, no division happening — real expertise, experience, past multi-domain decisions haven't been made."

'Founder mode'

Simo isn't a traditional Silicon Valley insider. Raised in Sète, a fishing town on France's Mediterranean coast, she was the first in her family to graduate from high school before earning a place at one of France's elite business schools.

From there, she worked her way into tech — first through an internship at eBay, then at Facebook, where she went on to help monetize the company's core app and eventually oversee some of its most ambitious product expansions. She also served as CEO at Instacart, where she helped steer the company through the pandemic boom and took it public in 2023 during a notoriously difficult market for tech IPOs.

Simo married her high school sweetheart, former software engineer turned chocolatier Remy Miralles, in 2011; they have a young daughter. Simo has spoken in the past about how navigating chronic health issues, including endometriosis and the nervous system disorder POTS, has shaped some of her work. She cofounded a women's health venture called the Metrodora Institute and was largely responsible for the launch of ChatGPT Health.

Former colleagues describe Simo as intense, empathetic, and known to crack a joke during a high-stakes meeting.

They say that background shapes how she leads. Sharma described her as a "hard-charging" executive with a distinctly human lens — someone who thinks about what ordinary people would actually find useful, not just what's technologically impressive.

Nick Turley, the head of ChatGPT who reports to Simo, said she has a "customer orientation" that has reshaped how the company approaches products. During an interview, he said that she has driven a new focus on reliability and performance over "flashy" new tech.

She also has a relentless operating tempo. Turley described it as her propensity to go "founder-mode."

"She will read every single document — including the links," he said.

Daniel Danker, who worked with Simo on Facebook's video team and later at Instacart, said "she was causing all of Facebook to move faster."

Simo's track record of commercialization is not without controversy.

At Facebook, she oversaw the company's video push, which later came under fire for inflating numbers. Facebook admitted in 2016 that it had overestimated how long viewers watched video ads by 60 to 80 percent; a lawsuit alleged the figure was as high as 900 percent. Facebook settled for $40 million while admitting no wrongdoing.

At Instacart, Simo inherited a fraught relationship with gig workers. On her first day as CEO in August 2021, she published an open letter pledging to be "a thoughtful and open partner" to the company's hundreds of thousands of gig workers, and invited them to email her directly.

The Gig Workers Collective, representing some 13,000 contract workers, called for a boycott and walkoff within weeks, telling Fortune that her responses were "basically canned answers."

OpenAI's next chapter

At OpenAI, Simo has helped push several high-profile initiatives, including a newsletter product called ChatGPT Pulse, OpenAI's Frontier enterprise agents, and advertising. This month, the company launched GPT-5.4, a model that incorporates coding capabilities into its core system, and announced a desktop "superapp," which Simo will oversee.

Simo said she approaches new product launches by involving employees throughout the process. She pitched those advertising roundtables by sharing her own ideas for what kind of ads the company could be proud of, and asking employees to share where they agreed and disagreed.

"The way I approached it was not to tell the company we're going to have ads. It's to actually start a dialogue," Simo told Business Insider. "That's not usually the way it goes. That's a big difference for this place."

As product chief, Simo has to prepare the company for battle inside a complex leadership structure, working closely with Altman.

Simo has said the two are complementary and aligned on vision, with Altman focusing heavily on research and compute scaling while she drives product execution.

It's a dynamic familiar to Simo, who worked closely with Mark Zuckerberg. She had a special talent for knowing how to navigate the CEO, Sharma said.

OpenAI can be a cutthroat place to work for leaders. As one former executive put it, OpenAI is such a rocket ship that there's very little time or patience for those who don't hit the ground running. "What a leader needs to do is astounding," the person said.

"The speed at which OpenAI is growing, it's relatively easy for Sam to hire the best, most famous people, but it's hard to keep them. His mode is: Nobody is not sacrificeable," the person said. "'You have to magically grow to what I think you can do right now.'"

OpenAI no longer has the luxury of being a research lab dreaming about the future of AI. It has rapidly become a global consumer product under intense scrutiny. Now Simo's job is to help it grow up without losing what made it successful in the first place.

Have a tip? Contact these reporters via email at astewart@businessinsider.com, gkay@businessinsider.com, or pdixit@insider.com. You can also reach them through the secure-messaging app Signal at +1-425-344-8242 and 248-894-6012, or via WhatsApp at +857-753-3949. Use a personal email address and a nonwork device; here's our guide to sharing information securely.

• Companies are spending big on AI, but research shows most haven't seen measurable ROI.
• To offset those costs, employers can do layoffs — or trim line items like bonuses and stock awards.
• Workers can still make a case for a pay bump by emphasizing their value and AI chops.

AI may be putting more than just jobs on the line.

Companies are investing heavily in the technology and top-tier AI talent, and layoffs aren't the only option for offsetting those costs. At least one survey of US business leaders suggests that some are planning to cut workers' compensation.

"They have to pay for AI somehow," said Rocki-Lee DeWitt, a management professor at the University of Vermont's Grossman School of Business. "It ain't cheap."

Research firm IDC says companies with more than 1,000 employees are expected to spend an average of $13.7 million on AI hardware, cloud infrastructure, software, and services this year, a 78% increase from 2025, based on a global analysis.

Nvidia CEO Jensen Huang said on a recent episode of the "All-In Podcast" that he would be "deeply alarmed" if one of the chip giant's top engineers spent too little on AI tokens. In another episode of the same program, venture capitalist Chamath Palihapitiya expressed concern about ballooning AI bills at his startup.

Yet despite all that outlay, 95% of organizations reported no measurable ROI from AI in the first half of 2025, according to an MIT study based on reviews of publicly disclosed AI initiatives and executive interviews.

Until companies see tangible gains from their AI investments, they may need to rein in other expenses, especially as tariffs, high inflation, and other factors also strain budgets.

Cost-cutting options

Several companies have announced layoffs tied to AI in recent months — including Block and Atlassian — and others may be tempted to follow suit, Business Insider previously reported. In November, HP said in an earnings report that it planned to cut between 4,000 and 6,000 jobs by the end of 2028, saving the company roughly $1 billion.

Cuts to employee compensation may be next.

More than half of 866 executives and senior managers polled earlier this month by ResumeBuilder.com, 58%, said they plan to reduce employee compensation by the end of this year to help fund their AI investments. Bonuses and stock awards will be affected the most, followed by raises, benefits, and base salaries, the findings show.

Though such cuts could hurt morale, "employees don't have any leverage" due to the tight job market, said Jessica Kriegel, chief strategy officer at workplace consulting firm Culture Partners in Sacramento, California. "They will push back less, and they will accept smaller raises to avoid risk that feels real."

ResumeBuilder didn't cite the size of the companies where the respondents work, though small businesses are the most likely to take a hammer to employee pay or perks in lieu of making job cuts, said Kriegel.

"You can't just lay off 10% of your organization when you have, say 20 people, and everyone has got their hands in a million different pots," she said.

Another option for companies grappling with AI and other costs can be to hold compensation steady. The Conference Board, a nonprofit provider of data and insights for business leaders, expects average salary increases to stay at 3.4% this year, the same as in 2025.

Getting a raise anyway

Workers can still make a case for a pay bump amid the AI boom, said Kris Erickson, cofounder of consulting firm Workforce Science Associates in Lincoln, Nebraska.

"When budgets are tight, and the economy is uncertain, you have to get into sales mode" to successfully secure a raise, she said. "You have to make yourself invaluable."

Given how much money companies are spending on AI in search of productivity gains, highlight any you've realized from using it. Merely saying you know how to use AI is not enough, said David Gaspin, an HR professional and executive coach in New York.

"Asking for a raise because you're proficient in AI is not the strategy," he said. "That proficiency is table stakes at this point."

It's also important to show why AI can't fill your shoes.

"The key differentiator today is becoming: What can you do that can't be replaced with technology? Where is your experience, your judgment, your unique point of view adding tangible, quantifiable value to the business?" said Gaspin. "Because that's where companies are going to see risk in you leaving."

• It's easy to dunk on Mark Zuckerberg and Meta for burning $80 billion on the metaverse and then moving on.
• But that's not exactly true.
• What is true is that Zuckerberg used to spend a lot of time talking about the metaverse. Now he talks about AI instead.

Nearly five years ago, Mark Zuckerberg told us the future was the metaverse — an idea that seemed to involve all of us strapping on virtual reality goggles and interacting with digital versions of ourselves.

Now, reports say Zuckerberg's Meta is bailing on the metaverse after losing more than $80 billion on the project.

This is a fun story for people who like stories about Big Tech tripping on itself.

But it's not really true.

Start with the $80 billion that publications like The New York Times and others say Zuckerberg has lost chasing the metaverse. Meta has indeed generated losses of at least $80 billion via its Reality Labs unit, which lost more than $19 billion in 2025 alone.

But Reality Labs is not going away. That's because Reality Labs makes lots of things beyond Horizon Worlds, the virtual reality space Zuckerberg told us that we would work and play in, but that almost no one actually visited.

Reality Labs also develops all the hardware Meta has been selling over the years, including its Quest virtual reality goggles, and its Ray-Ban AI glasses, which seem to have at least some consumer uptake (whether that's good for the world is a different issue).

At some point in the next couple years, Meta will roll out yet another set of glasses, purportedly designed to let you stream movies at home. (These are the same glasses Netflix co-CEO Ted Sarandos recently said director James Cameron can't stop talking about.)

It's entirely possible that all of Meta's device efforts will amount to very little. Efforts to get anyone but gamers to buy virtual reality headsets really haven't panned out, and while Meta, Apple, and others are now racing to bring the same tech to lightweight glasses, we have no idea if these things will ever be more than a novelty.

But for now, Meta is still plugging away at this stuff. Which means Reality Labs will continue to generate billions of dollars in losses this year and beyond.

OK. What about the idea that Meta is no longer interested in the metaverse — a notion Zuckerberg said was so important that he re-named his company after it?

That's a little trickier to assess. Meta is quite prickly about the notion that it's bailing on the metaverse: Its argument is that the metaverse doesn't have to involve headsets, and that you could do all kinds of metaverse-y things on your phone — or maybe your phone paired with some new glasses.

That's what Meta CTO Andrew Bosworth is getting at with this tweet he put out this week (and which Meta comms directed me to when I asked them for comment for this story):

Could be! But it's also true that Zuckerberg's public interest in the metaverse seems to have dramatically tapered off since 2021, when he told us the future was all about living in virtual space. (Zuckerberg had very different hair back then, too.)

Now, of course, Zuckerberg spends most of his time talking about AI, and Meta's ambitions to build "superintelligence." Which is why he's spending gazillions on AI talent and datacenters.

It's possible that all of those efforts get replaced by something else, too. Everyone in tech swears that the current AI boom really is a world-reshaping moment, and maybe it will be. But if you're still wondering what happened to all those NFTs you bought in 2021, I'll forgive you if you're going to remain in a wait-and-see on this one.

There is another way to think about Meta's interest in both the metaverse and AI. They're both shiny new things that offer Zuckerberg the promise of something he's wanted for a very long time: a way to run a business without having to rely on Google or Apple as his intermediaries.

Right now, Meta reaches people through phones and operating systems it doesn't control. At peak metaverse hype, Zuckerberg was clearly hoping to replace the iPhone with devices of his own. And in an AI-first world, it's possible the phone matters a lot less — or gets displaced by a new set of devices and interfaces.

That doesn't mean AI is just the metaverse with a new label. But it does suggest the through line here isn't the technology. It's Zuckerberg's recurring search for a platform he owns.

• Miro's CEO says the company is plowing cash into AI subscriptions to help employees level up.
• "Our L&D budget is unlimited tooling," Andrey Khusid said.
• AI adoption is accelerating, and with it come questions about the technology's ROI.

Plenty of companies are still debating whether costly AI subscriptions are worth it. Miro has gone the other way.

Andrey Khusid, cofounder of Miro, the maker of a popular online whiteboard platform, says the company gives employees essentially unlimited access to the latest AI tools as a way to speed up how quickly they learn and work.

That approach is possible, he said, because Miro has been profitable since 2016. The company has raised $476 million to date, and Khusid suggested it does not expect to need more capital.

Khusid framed the spending as a core part of more traditional workplace training. "Our L&D budget is unlimited tooling," he said.

Rather than asking employees to learn on their own time or pay out of pocket, he said, Miro wants that experimentation to happen inside the company, as a shared effort. He later added that there should still be a clear business case for buying any tool.

Miro's strategy is part of a wider shift in tech, where AI adoption is moving from optional to expected. A new study from engineering intelligence platform Jellyfish, based on data from more than 700 companies, found that 64% now produce a majority of their code with AI assistance. Tech giants like Google are pushing employees to use AI tools more aggressively, and Microsoft has begun tying AI usage to performance evaluations. As a result, AI fluency is quickly becoming a core workplace skill rather than a nice-to-have.

Still, Khusid says many executives ask the wrong question about AI ROI. Rather than judging the tools on individual productivity gains or subscription costs, he said Miro is trying to focus on whether the company is moving faster overall.

The company tracks projects through what he described as a "discover, define, deliver" process and measures how long it takes to move from one stage to the next. The goal is to compress that timeline as much as possible.

"The most important metric from my perspective is velocity of innovation," Khusid said. "If you don't innovate fast enough, you're out of the game."

Khusid said he doesn't think the way companies use AI today is necessarily the end state. He said it will take at least until the end of this year, or even next year, to see what a workplace shaped by these tools really looks like. At that point, Miro will take a harder look at which tools are worth the price tag.

For now, he said, Miro is already seeing time savings across engineering, product, and design. That's not always the case, though. Better tools speed code generation, he said, but code reviews can still bog down projects.

"Humans have to read it," Khusid said. At least for now.

Have a tip? Contact this reporter via email at mrussell@businessinsider.com or Signal at @MeliaRussell.01. Use a personal email address and a non-work device; here's our guide to sharing information securely.

• Srdjan Stakic vibe-coded a security camera system for his parents to ensure their safety.
• Stakic used vibe-coding platform Lovable to get started, as well as popular AI chatbots.
• His vibe-coded software became the basis for his AI-assisted startup Alvis.

This as-told-to essay is based on a conversation with Srdjan Stakic, 49, a former film producer who vibe-coded an AI system to monitor his elderly parents and detect falls. He's now launching a company that aims to offer the technology to others. This interview has been edited for length and clarity.

When I was diagnosed with stage four cancer two years ago, AI became essential.

Everything was happening so quickly: The doctors would talk to me for 15 minutes and leave me with more questions than answers. AI gave me an objective way to document and make sense of what was happening.

I'm now in remission. As my health improved, my parents' health declined, and I began helping them with cooking, cleaning, and medical appointments.

English was not their first language, and communicating with healthcare providers was tough. I recorded our conversations with their doctors and compared them to the after-visit summary using AI. I would put together all this information and translate it into Serbian for them.

But I soon wanted more than what the chatbots could offer. I wanted a system that could observe what was happening with my parents, or any other patient, and assess it through the lens of safety and dignity. I would think of how much guilt I'd feel if something happened and I wasn't there. What kind of son would I be?

I had never coded before, and I didn't have millions for an initial investment

I don't have a background in coding. I have a doctorate in health education and a master's in film production, and I have produced some films of my own.

I started outlining my idea with Gemini and ChatGPT to examine it from a tech and ethical standpoint. I built this document of what I wanted to achieve. I kept asking my family how they wanted to be treated in each scenario — like a fall or medical emergency — and I wanted to make the system flexible.

Then I transferred to Lovable. Lovable gave me a live development environment where I could describe what I wanted, see it built in real time, test it, and iterate. It connected the pieces, the frontend, the backend, the database, the authentication, the integrations, things I did not even know I needed until they were there. The chatbots helped me plan. Lovable helped me build.

I uploaded hundreds of training videos for nurses and healthcare providers to train the AI. I created a high-fidelity validation pipeline and a labeled dataset. I labeled real-world caregiving footage with established clinical benchmarks, like Stanford's C-I-CARE framework. When you approach a patient and introduce yourself, you tell them why they're there, you ask the patient's name and pronouns, and you introduce what you're about to do. You explain next steps and see if they have any questions or concerns.

I also started building an AI equipped with cameras to identify falls. I would fall in the middle of my living room and see whether the system recognized that and how long it would take.

It took me a few months to make it work

I tried different cameras and protocols, but ultimately, I had to hire an IT company to help me connect multiple cameras. The system can now identify a fall and send notifications to loved ones or EMS, and provide their location with a brief summary of their health records. The system also analyzes interactions between caregivers and my parents. It's sophisticated enough to analyze in real time — based on audio and video — if a caregiver is being rude or unprofessional. My parents have felt safer since I built this. I also built a feature that scans their environment for any trip hazards, such as cables.

I don't want to spy on my family, so I don't actively review all the video footage. When a concern is flagged, the system clips approximately 30 seconds around that moment and notifies mewith a summary of what it observed and why. It can also generate an advocacy letter from that same analysis: what was said, what was done, and how the interaction compares to the C-I-CARE framework to evaluate caregiver conduct.

I launched a company to offer this tech to others

This all started as an idea for my family, but the more I talk about it, the more people tell me they wish they had this for their parents. So I decided to launch a startup, called Alvis, to make this system available to others.

It detects falls in real time, recognizes when a caregiver goes above and beyond, and generates advocacy letters when something goes wrong. It's in private beta and accepting waitlist applications for our pilot cohort, launching April 13. The model will be a monthly subscription, similar to what families already pay for camera cloud storage, with a premium tier for AI-assisted analytics.

This week, my mom was hospitalized, and I used AI in four ways

First, I used it as a real-time medical interpreter: Every lab result went straight into Claude, so I understood what was happening immediately, not the next morning when a doctor was free.

AI was also my clinical advocate. When a history and physical exam understated her cancer history, Claude caught it. When her glucose started climbing from steroids, Claude flagged it.

Third, I used AI to translate updates into patient-friendly language in both English and Serbian.

Finally, Alvis — the camera system I designed — was running live in her hospital room all night, with her permission and a nod from her care team. It picked up her saying in Serbian, quietly, that she had endured too much. It flagged when I visited, and we recorded ourselves together.

It's amazing to see how vibe coding is democratizing access to AI tools. You can build a company that helps a very niche group that needs a specific thing. I still don't fully understand code or the extent of what I built, but it seems to be working.