Here’s an awesome tweet from Michael Truell, CEO of Anysphere, who make vibe code editor Cursor: [Twitter, archive]

We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week.

It’s 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM.

It kind of works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.

What an achievement! Three million lines is a large application. And Cursor built it completely autonomously! Allegedly.

This is Cursor using a framework with multiple chatbot agents. Imagine a slightly saner version of Steve Yegge’s Gas Town. The bot herder, Wilson Lin from Cursor, says: [Cursor, archive]

The agents ran for close to a week, writing over 1 million lines of code across 1,000 files.

The vibe code blogs, and some excessively gullible tech press, were bowled over! [e.g., Fortune]

So there’s only a minor problem with that Cursor announcement — every claim in it is a lie.

Cursor made one fatal mistake: they put the code up where other developers could see it.

The browser’s dependencies include html5ever, a component of the experimental Servo web browser; cssparser, another component of Servo; and a JavaScript interpreter that was already coded beforehand by Wilson Lin and included in the project by hand — not written from scratch by AI agents for this project. And a pile of other components that were absolutely not written from scratch either. [GitHub, archive]

It’s not clear how many of the long list of dependencies are used. Where the code doesn’t include other projects as dependencies, there’s a lot of what looks very like it was copied from existing code. Original theft! Do not steal!

None of this is written from scratch. The bot is putting together existing parts like Lego.

The supplied code for Cursor’s browser didn’t compile. When someone finally got it to work, it did indeed have rendering issues! The same rendering issues Servo has. But Servo is entirely in Rust, so that’s where GPT went looking for some Rust browser code it could use.

Servo maintainer Gregory Terzian was not impressed: [Hacker News]

The actual code is worse; I can only describe it as a tangle of spaghetti. As a Browser expert I can’t make much, if anything, out of it. In comparison, when I look at code in Ladybird, a project I am not involved in, I can instantly find my way around the code because I know the web specs.

What Lin and Cursor achieved was to show that an AI agent can generate millions of lines of code that’s lifted from other projects, and that don’t compile, let alone work.

Cursor’s fake browser is a marketing fraud to sell you vibe coding. This is standard. The guys selling you this trash will assure you it’s the best thing there’s ever been for coding. They will tell you impossible things that fail to check out when you spend a minute looking into them.

The vibe bros’ evidence for the power of vibe coding is “I feel faster.” They skip over the bit where it doesn’t work. They’ll make a big headline claim and try to walk it back when they get caught.

The target audience is CEOs who don’t realise these people are, to a man, brazen liars, and venture capitalists who want some hype to sell.

Never believe the hype. Always demand numbers on how effective vibe coding is. They all turn tail when you ask for checkable numbers.

• Video — Podcast

It’s pledge week at Pivot to AI! If you enjoyed this post, and our other posts, please do put $5 into the Patreon. It helps us keep Pivot coming out daily. Thank you all.

AI everywhere is — as you know — the future of business and civilisation! Except your company may have issues insuring against AI causing problems.

AIG, Great American, WR Berkley, and other insurers are asking permission from regulators to just … not cover AI. [FT, archive]

Berkley wants to exclude: [National Law Review, 2025]

any actual or alleged use, deployment, or development of Artificial Intelligence by any person or entity.

The list includes almost any way in which a business could touch AI. Berkley’s definition of AI seems to mean machine learning or generative AI:

any machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments, including, without limitation, any system that can emulate the structure and characteristics of input data in order to generate derived synthetic content, including images, videos, audio, text, and other digital content.

You can buy AI-specific coverage. Armilla offer AI coverage that’s remarkably difficult to actually claim on. But you can tell your boss you got coverage!

OpenAI and Anthropic have business insurance — but they both have multiple huge lawsuits in progress, which insurers don’t want to cover.

OpenAI is looking at putting investor funds aside into a special liability fund. Anthropic is paying its copyright settlement out of general company funds. [FT, archive]

If you want to pursue the fabulous AI future — have fun out there! But it’ll be your own backside on the line.

• Video — Podcast

With studies upon studies showing that actual measurable productivity gains through “AI” (which these days basically means chatbots) are really hard to come by and that “workslop” (meaning the extra work created for the rest of the organization by one person using “AI” lowering their work quality) eats up a lot of what might have been productivity gains many critics feel somewhat relieved: When the main narrative of “AI” (massively increased productivity) fails, this surely marks the end of this bubble and we might be able to get back to talking about actual problems.

I am not so sure.

Not because I think that some tweak to some LLM will make them suddenly that much more reliable or useful but because it was never about productivity really.

It’s something I alluded to in my “AI” talk at 2023’s Re:Publica: It doesn’t matter how good these systems are in reality, because that’s not what your boss cares about.

“AI” is a tool to disenfranchise labor.

That’s the job. If “AI” is actually more expensive than paying actual people actual wages that’s still a good investment for capital because it is about breaking up the structures, networks and organizations that help workers organize and fight for labor standards and fairer wages.

When CoCa Cola creates another bad ad using “AI” the fact that it’s garbage and expensive isn’t the point. It’s all just an investment into no longer needing to pay people for their expertise, work and time.

In December 2024 Ali Alkhatib wrote:

“I think we should shed the idea that AI is a technological artifact with political features and recognize it as a political artifact through and through. AI is an ideological project to shift authority and autonomy away from individuals, towards centralized structures of power.”

This is the macro-level view. It’s about Google, Microsoft and Amazon digging themselves even deeper into not only your personal life but also every economic workflow and process in order to charge rent using the established dependency.

But on a smaller (a bit short-sighted) level CEOs are looking towards “AI” not so much as actual replacement for labor but as the leverage to push down cost in the long term by crushing labor power.

So showing that productivity doesn’t measurably increase is winning the wrong game. It’s not an attack on our capabilities, skills or experience, it’s an attack on our collective and individual power.

The post Artist appeared first on The Perry Bible Fellowship.