I can't know everything. I have to trust someone else knows some parts and got it right so I can rely on them.

sometimes that trust is proven wrong. I have had to understand my compiler output to prove there was a bug in the optimizer (once I understood the bug I was able to find it was already fixed in a release I hadn't updated to yet). Despite that compilers have earned my trust: It is months of debugging before I think maybe the compiler is wrong.

I am not convinced that AI writes code I can trust - too often I have caught it doing things that are wrong (recently I told it to write some code using TDD - and it put the business logic it was testing in the mock - the tests passed, but manual testing showed the production code didn't have that logic and so didn't work). Until AI code proves it is worth trusting I'm not going to trust it and so I will spend the time needed to understand the code it writes - at great cost to my ability to quickly write code.