How do you keep AI-generated code maintainable as a codebase grows?

I've been watching teams ship AI-generated code faster than ever, which is great—until six months in, someone needs to fix a bug in a function that no human wrote and nobody fully understands.

The problem isn't that AI code is inherently unmaintainable. It's that when you're moving fast, it's easy to skip the review rigor that keeps any codebase healthy. AI-generated functions can work perfectly on day one and become a liability on day 180 when requirements shift.

I'm curious how people are actually handling this in practice. Are you enforcing stricter code review for AI contributions? Adding extra documentation? Refactoring generated code immediately into your team's style before it gets committed? Or does it depend entirely on what the AI was asked to generate in the first place?

What's been your experience—what's worked and what's backfired?