Letting AI review its own code is plain stupid

Forget AI for a moment and just imagine you have a mediocre engineer.

Not terrible. They can write code. They can ship things. But they're not senior, and their PRs regularly need improving.

A senior engineer reviews their work.

There are comments. Changes are requested. A few iterations happen.

Eventually, the PR gets to the standard you actually want.

So here's the obvious question:

Who thinks it would be sensible to put that same engineer in charge of reviewing everyone else's PRs?

I'd like to think there aren't many hands going up.

Because what happens? They review the code using exactly the same level of judgement they use when writing it. They miss things a stronger engineer would catch. They think the PR is good enough. Approved.

And that isn't really the engineer's fault. The fault lies with whoever decided to give them a responsibility they weren't capable of performing well.

Now bring AI back into the picture.

If I already know my AI-generated code needs review… If I already know its first attempt needs correction, refinement and iteration… Why would I then put AI in charge of deciding whether that code is good enough?

We'd call this bad judgement if we did it with people. Why are we pretending it makes sense because it's AI?

Who's at fault when that goes wrong? The AI? Or the person who thought removing the accountable engineer was a good idea?

Tell me where I'm wrong, or if you agree. Please reply in comments as we can't ignore this stupidity that's been dressed up as solving a problem.