Personally I still try to read or scan the generated code and look at the implementation and many design patterns are recognizable and are almost always exactly as how it should be done. I must admit that there are sometimes small parts of implementations that look inefficient or 'weird'. I remember that one time it wanted to store data in some sort of array type in a database instead of a separate table. Asking why AI did that it also agreed that it was a what tricky/shaky implementation. I didn't trust it and with a single prompt the code is rewritten into a perfectly fine separate table implementation.

That's the reason why I still try to inspect the generated code to see how my problem was solved and if it reads plausible. There are also moments that I am surprised by the implemented solutions which look very elegant and "smarter" than what I could have come up. These are also the moments I learn from AI. I can even ask AI why it did things certain ways and it gives a nice summary of arguments. But I am often also critical towards AI if I am in doubt, resulting sometimes in a sort of discussion and sometimes I can convince AI (with more context) that the solution it created should take another route. It really feels like working with a really fast co-worker. And on top of that, I am better in writing up my thoughts to discuss things instead of talking with someone. Doesn't mean I don't discuss in real life with co-workers, but it is so easy to use AI for it too.