When you do any meaningful work, that is, not "generate a website with a fancy UI", you very much realize that AI can not, in fact, "do the work". They constantly make mistakes, and you have to spend about as much time writing the spec and checking the code as you'd have written the code

So the effect is just merely some kind of acceleration of "boilerplate code writing", which is very impressive for beginner coders who are mostly doing automateable, trivial tasks, but much less so once you start doing real concurrency / threading / embedded / etc work