Anyone old enough to be in IT in the late 80's/early 90's will remember the "4th GL" phase that swept through Fortune 100 companies.

I was in banking at the time, and 'Focus 4GL' was brought in to replace programmers. Of course, in the end, it turned out to be a fools-errand.

My prediction though is that in less than a decade, ML/AI will be decent enough at developing solutions via client specs for many applications.

I originally got there by making a complex magic data structure that held relations to everything in multiple dimensions so I could generate a huge amount of stuff, but that turns out to be just like 4GL - a load of slow confusion. The reason I am doing it is exactly because of ML/AI hope - with enough data and proper structures, I can generate a lot.