According to Doug Slater, a developer writing on Hacker News, there's a paradoxical question worth asking: what's the optimal amount of low-quality, LLM-generated code you should ship? His answer: some. The key insight is that code review rigor should match actual risk. Your personal Python script needs less scrutiny than your company's authentication system, which needs far less than medical device software. Slater references a framework from software engineer Bertrand Meyer: casual software has limited use and loose constraints; business software can harm your organization; acute software can kill someone. Here's the trend he's concerned about: developers are skipping human code review entirely, even delegating verification back to the LLM that wrote the code. Slater himself has shipped unreviewed code—small macOS apps, personal utilities—because the downside is bounded. But for anything with wider reach or long-term maintenance, human judgment matters. His core concern isn't that LLM code is bad. It's that you lose your information advantage: your ability to understand the code deeply and know it actually works. Outsourcing that to a model that can't independently verify anything is, he argues, epistemologically unsound.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton