Really? Because I've seen the opposite: the cost of fixing AI code is significantly higher than fixing human-generated code, because step one of refactoring is

Really? Because I've seen the opposite: the cost of fixing AI code is significantly higher than fixing human-generated code, because step one of refactoring is understanding the code and verifying the tests cover all the behavior. And AI doesn't produce code optimized for any human to read it. Even the worst programmer understands their own code, whereas AI produces code no human has ever understood.