YC CEO Claims Daily AI Deployment of 37,000 Lines of Code, Review Finds Bloated Frontend

YC CEO Garry Tan claims he and AI coding agents deploy 37,000 lines of code per day across five projects. A developer audit of his personal site frontend revealed extensive bloat instead of clean output.

The claim first appeared on X. Tan posted that the 72-day streak continued without interruption. He presented the volume of output as proof that agentic AI can accelerate product work.

A Polish developer named Gregorein examined the live frontend code of Tan's site. The page made 169 requests and transferred 6.42 MB of data. By comparison, the Hacker News homepage uses seven requests and 12 KB total.

Gregorein listed concrete issues. The bundle contained 28 test files that remained in production. Seventy-eight JavaScript controllers showed no active references. Eight different logo formats existed, including one empty file. Several old PNG assets were left uncompressed.

These findings illustrate the core tension. High line counts do not equal production quality. AI systems can emit large volumes of code quickly, yet the resulting artifacts often require heavy human cleanup to reach acceptable performance.

Tan has not responded publicly to the audit. The post remains visible on X. No independent verification of the 37,000-line daily figure has been released by Y Combinator or any third party.

Other YC founders have discussed AI tooling in public posts. Most emphasize review steps after generation. None have published comparable daily line-count metrics.

The event places pressure on the broader agentic AI narrative. If volume alone becomes the headline metric, teams may accept bloated bundles that slow down real users. The review shows one concrete case where quantity arrived first and efficiency followed later.

Gregorein concluded that quality must remain the priority. The audit serves as an external check on unchecked generation speed.

Future signals worth watching include any public reply from Tan or changes to the audited site assets. Additional developer audits of other founder sites could reveal whether the same pattern appears elsewhere. Metrics on actual page-load times after any cleanup would indicate whether the bloat was addressed.

Developers evaluating agentic tools should track both output volume and the size of resulting bundles. The Tan case supplies one data point that volume claims alone do not guarantee lean production code.