The issue with current AI train hype in general is the gap between a cool demo and thinking that it means you can prompt anything just because the demo worked. This is really obvious with images or video. There are some amazing examples, beautifully rendered, and it gives the impression you can prompt anything that comes to mind and it will work. The reality is most prompts don't work as expected. Many concepts don't work as expected. There is a space of "viable prompts" that produce acceptable outputs and anything outside of it is unusable. Similarly with code there is a "space" of languages, projects, features, etc., that will produce decent output, but of course the vast majority of things that are possible and novel are not in this space.