Garry Tan, President of Y Combinator, claimed he generated 37,000 lines of code in a single day. That's wild. A productive developer writes a few hundred lines of production code daily. Developer Gregor went and audited Tan's website to see what that output actually looks like.
The code is public. Draw your own conclusions. But notice what nobody in this conversation is asking: is it any good? Cantrill argues that LLMs lack the programmer's real virtue: laziness, the drive to create efficient abstractions. He cites Tan's claimed 37,000 lines per day as an example of how AI tools enable a 'brogrammer' mentality of generating massive amounts of low-quality code.
Tan co-founded Posterous, got acquired by Twitter in 2012, and built his career as a software engineer before running Y Combinator and Initialized Capital. He knows code. He also knows lines of code was always a garbage metric. So why brag about the count?
AI coding tools can dump thousands of lines of generated code in minutes. Speed was never the bottleneck in software. Figuring out what to build and making it maintainable is where the actual work happens. A line count tells you nothing about either.