Or is the bad code you've seen generated by others pretty terrible, but the good code you've seen generated by others blends in as human-written?
My last major PR included a bunch of tests written completely by AI with some minor tweaking by hand, and my MR was praised with, "love this approach to testing."