Having learned many of the same lessons as you I can 100% backup everything you said in your “sum it up” paragraph!

The only caveat I would add is to the “make an MVP with AI”. I think MVPs generated directly out of ChatGPT/Claude are so easy now (or at least it can appear so on the face of it) that many people are just barely going beyond that - but to any experienced eye, that approach is quite transparent and can look very low-value (even if the idea is actually a good one).

Now if that person is a skilled salesperson then that might work.

But, for most people, I think it’s still very important to demonstrate good instincts, taste and strategic/commercial understanding when building such an MVP. And that means editing and shaping the output just enough to meet your vision for the product. So to agree with you - definitely, 100%, use AI as much as possible - but don’t assume that you can put zero work in on top and have that MVP be effective. Because the 10 year old down the street has the exact same tool as you - so if you are just relying verbatim on that tool’s output- it’s going to be hard to stand out.

I’d still definitely agree to spend as little time as physically possible on the MVP - with the above caveat.

Having said all that… a lot of historical wisdom on the topic of MVPs has been turned upside down since gen AI became mainstream, so on the flip side you could argue: create 1000 MVPs in an hour, publish them all, see what generates interest…*

Hmm.. I think I just argued against my own point.

* (I’m not really seriously suggesting anyone do this, but I’m also not entirely discounting this as an approach either…)