The code had tests, which is the real-world check against bad code, both AI and human written.

Good code is not code that “just works“ is readable code.

Tests is just one part of QA. Code review is another.

Except for the pesky turtles nature of that logic..

All the way down...

You can't fix it.

You ask the AI to fix it and hope, or you start again from scratch - which if an AI is making the code, might be just fine.

But I think you still need a type of task AI can do well - something which lends itself to auto-complete.

Actually, if the AI is making the code, starting from scratch might not be great. You'd get a new and different set of bugs.

It's hilarious how the discussion in the PR is not about the PR itself, but about the implications of this.