Good is loaded. When given a prompt that sets scope well, AI more often than not now writes correct code which by most programmers' definition is "good" ... enough. AI still does not write code that's as a clear as most "good" human programmers. For people that don't read their own/other people's code that's probably okay, and AI probably does write more readable code than they do because they never developed a taste for it.

AI doesn't write my code. It may prototype something for me, or fix up something I messed up, but it rarely leaves my private env.

My #1 action is: ask bot to write PR internally, look at the code, meathook it one abstraction level higher, let bot review 20x but with comments, not code, then fix all that up into something neat.

Lot more work, 10x more readable code as a result.