When AI 'builds a browser,' check the repo before believing the hype

AI “built a browser”? Devs checked the repo and LOL’d

TLDR: Cursor hyped an AI-built browser, but developers who cloned the code found it barely runs and leans on existing tools. The community roasted the line-count brag and marketing spin, with some defending it as a demo—ending in a clear takeaway: don’t trust the hype, check the repo.

Cursor’s big flex—“We built a browser with GPT‑5.2”—hit social feeds like a rocket, with CEO Michael Truell touting 3 million lines of code and a “from‑scratch” engine. Then he added it kinda works, and the internet did what it does best: cloned the repo and hit play. Cue the record scratch. Reviewers say it barely compiles, needs manual patches, runs painfully slow, and leans on existing projects despite those “from scratch” vibes. The engineer’s own blog post admits building a browser is brutally hard—yet the marketing read like a victory lap.

Drama exploded. On Y Combinator, Wilson Lin defended parts of the JavaScript engine as a vendored personal project, which didn’t help the “AI wrote it all” headline. A Servo maintainer dropped the week’s harshest burn, calling it “a uniquely bad design”—the kind of quote that gets screenshotted into oblivion. The crowd clapped back at the zillions of lines brag: quantity isn’t quality, especially when your Continuous Integration (the automatic tests that prove it works) is failing.

Hot takes piled up. One commenter started supportive, then posted “EDIT: I retract my claim.” Another lobbed “grifters gonna grift.” Someone even declared AI can’t beat chess grandmasters—instantly memed, because… it famously did. The mood? Hype vs. reality, with devs chanting the week’s lesson: check the repos before believing the sizzle.

Key Points

  • Cursor claimed its AI agents built a web browser with a from-scratch Rust engine and custom JS VM, generating over 3 million lines of code.
  • Wilson Lin’s blog post showed only limited functionality and acknowledged that building a browser from scratch is extremely difficult.
  • Developers reported the repo barely compiles, CI (GitHub Actions) fails, and no recent commits build cleanly.
  • Performance where builds succeeded was poor (about a minute to load pages) and relied on dependencies like Servo and QuickJS.
  • Estimates via Perplexity suggested the experiment used 10–20 trillion tokens, potentially costing several million dollars at list prices.

Hottest takes

“EDIT: I retract my claim” — jey
“I’m super impressed by how ‘zillions of lines of code’ got re-branded as a reasonable metric” — Sharlin
“grifters gonna grift” — blibble
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.