The code that LLMs produce is just average IMO. I wouldn’t call myself an authority on clean code but I can tell when code is well structured. I prefer my hand

The code that LLMs produce is just average IMO. I wouldn’t call myself an authority on clean code but I can tell when code is well structured. I prefer my hand written code over Claude or GPT’s every time. I once did an experiment where I generated a spec from a project I’d already written, then had an LLM blindly reimplement it from the spec, and compared code. The LLM’s version looked like vomit.