Humanity suffered „reasoning by correlation“ for millennia, you can’t go that far with such „fundamental flaw“.
These things help with prototyping at snippet-level, more than that is just a silly game.
Yes, it is not sustainable at this scale.
ML is useful (Recognition, NLP), not for precision-works

Europe is awesome! semaphores, mutexes, null pointers... switches 🤭

It depends on the attitude and the things you are working on.
AI tools align perfectly to the general business attitude of getting things done as fast as possible with acceptable results.
Engineering and scientific attitudes are radically different.
Plus, review/ correcting code is more difficult.