Completely agree, these are good points. I'll add that current AI is overhyped precisely because it is not 100% reliable, especially with novel or cutting edge

Completely agree, these are good points. I'll add that current AI is overhyped precisely because it is not 100% reliable, especially with novel or cutting edge problems (where the most utility exists). This means that engineers using AI ultimately have to understand everything that the AI is doing, and this reduces its utility because you still need to be as much as an expert as the AI, if not more. It's similar to self-driving cars and how they require the driver to have eyes on the road and hands on the wheel. Sure, it is a useful feature for drivers just as cruise control was, but it is not going to free someone from the task of driving. That said, self-driving is getting better and better, but is it realistically nowhere near the point of allowing someone to fully trust the car and therefore free themselves from paying attention to the road. Sooner or later you need to take control of the car and so "self-driving" is merely a marketing term that is overhyping what is effectively an enhanced form of cruise control (even self-driving taxi services still require remote operators). Similarly, AI for coding will require engineer expertise and supervision, and it might restructure how teams do work but it will not fulfill the promise of letting the AI code full solutions and rely on the solution to not break.