AI coding speed can hide messy, bloated code
Y Combinator CEO Garry Tan said he and his were shipping 37,000 lines of code per day across five projects, with a 72-day streak of releases. Developer Gregorein inspected the public front-end code of Tan’s AI-focused site, garryslist.org, and found that the fast output came with heavy waste. Loading the homepage made a browser request 169 files totaling 6.42 megabytes. Hacker News, also run by Y Combinator, was used as a comparison point with seven requests totaling about 12 kilobytes.
The site sent 28 test files to every visitor, even though visitors do not need them. It also loaded 78 for features that did not appear on the homepage. The logo was downloaded in eight formats, including one empty 0-byte file. Two large PNG images used about 4 megabytes, though newer image formats could have cut them to around 300 kilobytes.
Other issues included duplicate page content, an empty CSS file, a large editor loaded on a read-only page, missing image descriptions, and analytics code routed through a proxy to avoid ad blockers. The review only covered browser-visible front-end code, not the back-end or database, but the main lesson is clear: AI can create code faster than people can comfortably review it, so quality checks matter more, not less.
Key points
- Garry Tan said helped ship 37,000 lines of code per day across five projects.
- garryslist.org loaded 169 files totaling 6.42 megabytes on the homepage.
- The site included test files, unused , oversized PNG images, and duplicate or empty files.
- The review covered only front-end code, not the back-end or database.
- AI coding can increase speed, but unchecked output can create slower pages and harder maintenance.