[contextfree]

In the broader context, you could look further left:

Conception -> design -> compiler -> code review ...

If AI tools allow for better rapid prototyping, they could help catch "errors" in the conception and design phases. I don't know how useful this actually is, though.

[20k]

One of the problems with using AI for prototyping (or just in general), is that the act of creating the prototype is what's valuable, not the prototype itself. You learn lessons in trying to build it that you use to build the real product. Using the AI to skip the learning step and produce the prototype directly would be missing the point of prototyping at all

[contextfree]

That's definitely an issue and I've gotten burnt by similar problems when using AI to help navigate and find things in codebases, where I've used it to read and understand code for me, with seemingly miraculous results at first, but ended up wishing I'd read more code "manually" myself, as my shallow understanding led to wasting more time on net than I saved. I still feel like it should be possible to find some kind of a balance, but it's tricky.

[amilios]

Sometimes you need to just pump out an MVP and start iterating. AI can exponentially speed this up in my experience.