Summary

Dan Luu argues that blindly rubber-stamping LLM outputs — acting as a 'meat proxy' or 'for loop' — is a self-defeating strategy for software workers. As LLMs improve toward the point where they could handle routine development with minimal oversight, the worker who provides no judgment offers nothing a company couldn't get by running the LLM directly and cutting the headcount. He marshals empirical evidence — broken GitHub repos, buggy commercial products, board game AIs weaker than naive humans — to show that 'coding is solved' claims are demonstrably false today, and that out-of-distribution problems still require genuine human judgment. The conclusion is that regardless of whether AI eventually displaces programmers, turning off your brain is wrong in every plausible future: it produces poor work now, and it guarantees obsolescence the moment LLMs get good enough.

Key Insight

Acting as a mindless LLM executor is uniquely self-undermining: the better LLMs get at the tasks you're proxying, the less reason there is to keep you in the loop at all.

Spicy Quotes (click to share)

• 8

  There's no point at which this methodology will work for the employee.

• 8

  In everyday changes, my reviews often cut the diff to 25% of its original size. Tons of useless tests; paranoia; inverted logic. Then I read Twitter and 'coding is solved'.

• 5

  as more evidence has come in, I've gotten more sure that it's just that people are fooling themselves.

• 6

  if your goal is to do as little work as possible, the best time to do this was in the past.

• 7

  whenever I say 'LLMs don't solve all programming problems' ppl look at me like I'm crazy, and I look at them like they are.

Tone

analytical, critical