• A trio of journalists tried three big vibe coding apps to see how they stack up.
• We each attempted to build an app on Cursor, Lovable, and Base44.
• With the same prompt on each system, we wanted to see how far we could get.

We three writers have been handed a gift with seemingly infinite potential. A sparkling promise, from vibe coding startups, that we can build anything without understanding a word of code.

Gone are the days, these companies say, when coding novices needed to rely on their techie friends to troubleshoot mistakes.

Over a dozen firms have rolled out tools offering the ability to build apps in seconds. All you need is a good idea and the platforms' free coding credits.

With a growing wave of vibe-coding startups raising big money, questions are emerging: Are these tools meaningfully different? Is the market already crowded? And can this be a sustainable business?

We tried three of the most popular platforms — Cursor, Lovable, and Base44 — to find out what each platform really offers and where they fall short.

Our prompts

We started this experiment at different levels of proficiency.

Chong Ming had coded an app at a vibe-coding workshop. Cheryl, self-taught, had experimented with five vibe-coding platforms and made three web apps. Aditi was a true beginner.

We each set out to make an app. For Chong Ming, a writing companion in the shape of a cute creature. For Cheryl, a newsroom dashboard, a lite version of Asana to keep her team's work organized. And for Aditi, an app that acted like a newsroom photo coach, to deem whether a photo was good to publish.

First impressions

Chong Ming: When I asked Base44 to plan the app, it responded with a few questions to clarify my prompt, with cute emojis. The plan it generated wasn't as detailed as what I've seen from Cursor, but it was user-friendly.

Lovable's plan was even simpler, pared down to a few bullet points. It was just as easy to use as Base44. Within minutes, it generated a web app similar to Base44's.

Cursor's interface seemed built for serious builders looking to ship real products. Its planning questions were more advanced and thoughtful, asking if I wanted an MVP build plan, a clickable prototype, or a full product spec — the kind of distinctions a software engineer would make.

Cheryl: I've always been one for the rule of cool, and Cursor looked really cool, with its all-black dashboard. But there was some charm in logging onto Lovable, with its girlypop, pink-heavy interface, and Base44 offered some cheerful vibes on its tangerine-colored interface, too.

Base44 and Lovable felt more like signing into a website and conversing with a chatbot. With Cursor and its MacBook app, I felt like I was hacking into the mainframe, with all its complicated scrolling lines of code.

Aditi: Off the bat, Cursor looked intimidating, and the option to sync GitHub when logging in made me think it wasn't a platform for a non-technical user.

Meanwhile, Base44 and Lovable were friendly and reassuring, with their gentle prompts: "What will you build next?" and "Ready to build, Aditi?"

Learning curve

Chong Ming: Base44 and Lovable were easy to use. The app plans were written in plain language for everyday users, and the interface was beginner-friendly. It was clear where to click if I needed help or wanted to tweak something.

Cursor was a different story. There were things I had to decipher on my own, like "frontend built with Next.js, React, and TypeScript."

Cheryl: I have never felt more like a dinosaur than when I first tried using Cursor. It was embarrassing having to look up basic terms to know what I was dealing with.

On Base44 and Lovable, I consistently typed in plain English and made the app edits accordingly. I felt like a wizard, watching the app preview morph and shift into view.

Aditi: I'd never tried vibe coding, so I asked AI for help understanding AI. I asked ChatGPT to help me refine my initial prompt into something I could plonk into the vibe coding platforms.

With Lovable and Base44, the learning was intuitive, and it felt like I was talking to ChatGPT. With Cursor, I was completely lost and had no idea where to start.

Then it was time to build the apps

Chong Ming: Base44 built me a writing companion app with a cute egg. The layout felt bland, but it was a full-fledged, functional app created without using up all my credits.

Lovable's build was similar and didn't use all credits.

Both platforms could generate the app; the main variation was in aesthetics. I did appreciate that Base44 and Lovable let me edit the app directly in the interface.

The Cursor build process wasn't as hands-off. Unlike Base44 and Lovable, which ran start to finish, Cursor required me to approve commands and grant permissions to override folders on my computer. As it generated code, I could pause and review it, something that would likely appeal to developers who want control.

Cheryl: The best things in life are free, and vibe coding credits are one of them. On Base44 and Lovable, both platforms make it clear to users that they're cooking with limited credits, and that's fair — compute is costly. The mileage on each platform, however, was slightly different.

Lovable gave me good bones for the project up front and created something that was, aesthetically and functionally, closest to what I wanted. But it burned through more free credits than my Base44 project did, and some things still weren't working in the web app. I was stuck waiting for new credits to drop before I could make tweaks.

Base44 gave me something very close to a complete dashboard, but it lacked some key functionality — the option to delete tasks, or to drag and drop unscheduled tasks into the calendar frame. But that was ironed out within minutes with two additional message prompts.

Cursor's steeper learning curve and multi-step process made it far harder for me to work things out. After 10 minutes of Googling, I gave in and typed into the Cursor chat: "I'm confused. What do I do now? Give me a guide."

I was told to go to Supabase and make some adjustments to the settings, then try to ship it via a local server. At that point, I was coming up on half an hour of getting frustrated with the process.

Aditi: The development process was smooth sailing with Lovable and Base44. With one initial prompt and two additional tweaks, both platforms gave me usable apps that I thought would be handy newsroom tools.

I first tried Base44 and felt childlike wonder when it produced a clean, minimalist interface that let me drag a photo in and judge its quality.

After the initial merriment wore off, I started testing the features. One thing I had not realized was how specific my prompts needed to be, expecting it to anticipate my needs. For example, both platforms initially did not allow me to crop the image or adjust the framing, and instead automatically chose the subject for me. An easy second prompt brought the apps closer to my initial vision — although I quickly learned how to ration my prompts lest I run out of my daily free credits.

Lovable's interface had a neat little photo-scanning animation that I thought added visual interest to the otherwise simple interface.

Now for Cursor. I had to download the app on the MacBook, while the others could run in the browser. When I finally downloaded it and fed it my prompt, it ran lines of obscure code, asked for permissions to things I didn't understand, and made me lose motivation to build anything.

I eventually gave up on trying to make it work, but the app kept prodding me with pop-ups for permissions all day until I force-quit it.

I'll stick to my beginner-friendly platforms until further notice.

How the platforms stack

Chong Ming: Lovable and Base44 delivered working apps and refinements fast, but the quality didn't match Cursor. Cursor broke down what it added and made changes in detail, even if some of the jargon flew over my head.

When I refined the app, Cursor didn't just tweak surface-level things. It suggested enhancements such as adding extra animation frames or making the pet move faster. When I said I didn't want a simple egg, it flagged that a drawn mascot or pixel pet would require new assets — a level of clarity the others didn't offer.

By comparison, Lovable and Base44 suggested things like adding entrance animations, which felt more gimmicky than meaningful.

If I were building something serious, I'd go with Cursor, even if it takes more time and effort to get up to speed.

Cheryl: On both Lovable and Base44, I managed to build workable newsroom calendars and get them from first prompt to publishing within 10 minutes. Base44 gave me a complete, fully functional project I could immediately use and share with my team — and within the free credit range, too. The next day, I used my new set of credits on Lovable to make final tweaks, resulting in a publishable dashboard with all the functions I wanted.

On Cursor, however, I just couldn't figure out what I was doing wrong and why the code wasn't running as I intended. I never got my dashboard off the ground there. Cursor: It's not you, it's me.

If you have a nontechnical background, a clear vision for the app you want to build, but limited time to pick up a little more coding, a one-stop shop like Lovable and Base44 would be more your speed. If you do have more coding know-how, Cursor will give you access and oversight over the coding process within its free credit limit.

Aditi: As a colors-obsessed, minimalism-loving, non-technical person who just wants to build a simple app, here's my leaderboard: Lovable, Base44, Cursor.

The market's flooded with options, so take your pick while companies are being generous with credits

The apps we tried are just a sampling of the vibe coding offerings out there. Other companies, like Emergent and Replit, also offer one-stop-shop platforms that take ideas from conception to shipping fast.

The barrier to entry is low, particularly with free credits on entry-level plans.

So if there was ever a time to try vibe-coding, it's now.

• 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.

• Amid China's OpenClaw craze, the AI agent is now moving into robots.
• China's tech giants have begun launching their own versions of OpenClaw in the past weeks.
• Meanwhile, the US is still concerned about AI agents going rogue.

While much of the world is still experimenting with OpenClaw, China is already putting it into robots.

Chinese home robotics giant Ecovacs unveiled its new robot, Bajie, powered by OpenClaw, at a consumer electronics expo in Shanghai last week.

Advertised as a home "butler," the robot can perform household tasks such as tidying shoes or putting away toys.

Ecovacs founder Qian Dongqi said in an interview with Chinese outlet Ifeng that the long-term goal is for robots like Bajie to take on more household chores.

A writer from the Chinese tech outlet 36Kr who saw the robot in action reported that it required multiple prompts to complete tasks and "there were also unstable situations."

It's not just home robots. Developers have begun integrating OpenClaw into Chinese robot-maker Unitree's G1 humanoid robot, allowing it to interpret commands and navigate physical spaces in real time. A US-based team, Dimensional, has open-sourced the system behind these integrations.

Another Chinese company, AgileX Robotics, earlier this month published a guide showing how OpenClaw can be integrated with its robotic arm, letting users control the machine through natural language.

Chinese tech giant Xiaomi is also testing its version of OpenClaw across its ecosystem, from smartphones to smart home devices.

China has been gripped by an OpenClaw craze lately. Users rushed to install the AI agent on their devices, with some paying strangers to set it up for them and others forming long queues outside Tencent's Shenzhen headquarters and Baidu's Beijing office to get help from engineers.

The OpenClaw obsession is partly driven by the viral phrase "raising the lobster," which Chinese users use to describe deploying the AI agent to automate everyday tasks.

To meet the demand for AI agents, China's tech giants, including Tencent, Alibaba, and ByteDance, have begun launching their own versions of OpenClaw in the past few weeks.

US concerns about security

Meanwhile, in the US, concerns about AI agents going rogue continue to grow.

Last month, Meta's alignment director, Summer Yue, connected OpenClaw to her inbox, and said in an X post that the bot tried to delete her emails.

"I had to RUN to my Mac mini like I was defusing a bomb," Yue wrote on X.

In a separate incident, an AI agent set off a major internal security alert at Meta after acting without approval, exposing sensitive company and user data to staff who weren't authorised to see it, The Information reported on Thursday.

Tech leaders have also sounded alarms. Elon Musk last month posted an image of a monkey being handed a rifle on X, captioning it: "People giving OpenClaw root access to their entire life."

Even Nvidia CEO Jensen Huang, who has praised the technology, has emphasized the need for stronger safeguards. His company is working on its own agent system, NemoClaw, with a focus on security.

• China's biggest tech names all want a piece of OpenClaw.
• Tencent, Alibaba, ByteDance, and others have launched versions of the AI agent and integrations.
• OpenClaw took China by storm in recent weeks as the phrase "raising the lobster" trended online.

The "lobster" craze in China has quickly turned into a corporate land grab.

Within weeks of OpenClaw gaining traction among developers and hobbyists, China's internet giants began rolling out their own versions of the AI agent and integrations.

Tencent launched QClaw last week, a tool that integrates OpenClaw into WeChat's vast ecosystem, China's super app. Users can send a message directly to QClaw via WeChat, and the agent will immediately execute the task, Tencent said on its website.

TikTok owner ByteDance's cloud unit, Volcano Engine, rolled out ArkClaw, a cloud-based version of OpenClaw accessible through a web browser. Alibaba also introduced JVS Claw, a mobile app designed to help users install and deploy OpenClaw more easily.

Xiaomi, a consumer electronics company, has launched a closed beta test of MiClaw, an AI agent that lets users control Xiaomi smartphones and smart home devices with single-sentence commands.

AI startups moved just as fast. Zhipu AI, Moonshot AI, and MiniMax have released large language models or frameworks built on top of OpenClaw. Shares of Zhipu AI and MiniMax surged 13 per cent and 22 per cent respectively last Tuesday, following the announcements of their OpenClaw tools.

It's not just Chinese companies. Nvidia on Monday announced that it has created NemoClaw, an enterprise platform built on top of OpenClaw.

"It has a network guardrail, it has a privacy router, and as a result, we could protect and keep the claws from executing inside our company, and do it safely," CEO Jensen Huang said during Nvidia's 2026 GTC conference in San Jose on Monday.

"Every company in the world today needs to have an OpenClaw strategy, an agentic system strategy," he added. "This is the new computer."

OpenClaw has taken China by storm. The trending phrase "raising the lobster" went viral, as Chinese social media users used it to describe deploying the AI agent to automate everyday tasks.

People across China also rushed to install OpenClaw on their devices, forming queues outside Tencent's Shenzhen headquarters and Baidu's Beijing office to seek help from engineers. Others turned to online marketplaces, paying strangers to install the tool for them.

The frenzy has since been tempered by growing security concerns. In the past week, some users have begun removing the software — in some cases, even paying others to uninstall it.

Earlier in February, China's National Vulnerability Database, run by the Ministry of Industry and Information Technology, warned that the open-source agent could introduce security risks if not properly configured. Misconfigured deployments could leave systems exposed to cyberattacks or data breaches, it said.

Last week, Chinese government agencies and state-owned firms moved to curb the use of OpenClaw on work devices.

Do you have a story to share about tech in China? Contact this reporter at cmlee@businessinsider.com.