Should AI-generated code be verified by another AI?

I've been thinking about this a lot while building CODYGIENE.

AI coding agents are getting extremely good at generating and modifying software.

But there's something about the current workflow that bothers me:

AI writes the code.

AI reviews the code.

AI fixes the code.

AI says the code looks good.

That creates a very convenient loop, but not necessarily an independent one.

My current view is that code generation can be probabilistic, while as much verification as possible should remain deterministic and reproducible.

That's one of the ideas behind CODYGIENE.

It runs deterministic static analysis across security, performance, code health, AI quality and structural architecture signals, locally on the repository.

The AI can still help explain or fix something afterward.

But it shouldn't be the authority deciding whether the problem existed or disappeared.

I'm curious how other builders see this:

Would you trust an AI agent to be the primary reviewer of code it generated?

Or do you think independent deterministic verification becomes more important as coding agents become more autonomous?