fastcompany.com via Hacker News

Developer audits YC CEO Garry Tan's 37K AI lines-per-day claim

TL;DR

  • Y Combinator CEO Garry Tan posted on X that agentic engineering is letting him ship 37K lines of code per day across 5 projects on a 72-day streak.
  • A Polish developer using the handle Gregorein audited Tan's public site and reported 78,400 lines of code, with a single homepage load pulling 6.42 MB across 169 requests.
  • Findings included 28 test files served to visitors, a logo in 8 formats with a 0-byte AVIF in production, and a 520 KB rich-text editor on a read-only page.

A Y Combinator CEO bragging about output on X collided this week with a line-by-line technical audit, and the audit is the part getting shared, according to Fast Company. Garry Tan had posted that agentic engineering was letting him ship 37K lines of code per day across 5 projects, still speeding up, on a 72-day shipping streak. A Polish developer using the handle Gregorein pulled the code apart and posted the findings on X.

The headline number in the audit is that a single homepage load of Tan's site downloads 6.42 MB across 169 requests. The comparison Gregorein reached for is Hacker News, which Y Combinator also runs, at 7 requests and 12 KB. Under the hood he reported 28 test files being served to every visitor, 78 Stimulus controllers downloaded for a page that does not use them, a logo shipped in 8 formats (including a 0-byte broken AVIF that made it to production), article images of 2.07 MB and 1.99 MB as uncompressed PNGs, a 520 KB rich-text editor loaded on a read-only page, and 47 images with empty alt attributes. The whole surface he put at 78,400 lines of what he called AI slop code.

The stakes here are not really one CEO's personal blog. The developer quote that Fast Company picked up is the one that lands, that AI lets you generate code faster than any human can review it and the answer from people like Garry seems to be so stop reviewing. That is a governance question for every startup being told to copy the playbook, and for every product on the receiving end of an agentic PR stream.

The honest caveat is that this is a single hostile audit of one public-facing site, and a raw line count is easy to inflate with generated boilerplate, tests, and deletions. What the reporting does not give you is whether Tan runs any linter, CI, or human-review gate behind the agent, or whether Y Combinator plans a response. The upside for anyone selling automated review, web performance tooling, or old-fashioned code craftsmanship is that lines-per-day just became a much less credible vanity metric.