It's trained on massive amounts of public code, so its output is (at best) average-- like the code it was trained on. My issue isn't really the output itself, but the mindset that accepts average output as good enough. People who make that choice will keep producing below-average software, and that's what the rest of us will have to live with.

You're being silly. With an appropriate dev pipeline it comes out quite nice. A team of agents to design and plan the implementation - a team of coders - and a team of reviewers after the fact.

Here, let's test it. Give me some of your code (GitHub is fine) and we can compare.