I think it's more that unlike with developers in general, the statistical distribution of errors in AI generated code is Uniform Random, meaning that unlike with human produced code reviewers can't just concentrate in checking for certain kinds of mistakes to catch most stuff, they have to dive equally deep on everything and, yeah, that includes checking for stupidly basic mistakes since those are just as likely for AI to make.

This also means that automated testing (such as unit testing) has to cover everything, not just focus on boundary conditions and corner cases - AI can just as easilly create bugs in "nobody would get this wrong" domains as it can in highly complex ones.

This is yet another way in which AI coding just moves the manpower costs from the code creation phase to other phases of the software development process and overall application life-cycle, rather than reducing overall costs.

I would even say that for any software that has a full life-cycle (i.e. pretty much anything that needs to be maintained and upgraded when in Production, so anything but one-off "code it and push it out the door" projects) it seems to be increasing overall costs.