People tend to forget that half of all programmers are below average. The amount of human slop that is generated by human coders is immense. I know because I've worked in some of those companies, some of them even YC startups. Human-generated code is not sacred, it's mostly garbage that barely works.

So when articles talk about how bad AI-generated code is or it doesn't understand how to design things properly... this is how most human-generated code is! In fact I would bet that AI-generated code from the frontier models are consistent and pretty good, at least the stuff I've seen. The difference is that you can just tell the AI to rewrite it and it will do it in seconds. And it will only get better as time goes on, and you can send it to back to older code, tell it to rewrite it without losing any of the existing nuances and then add test code and it will in seconds or minutes.