It's an old practice - whether you call it a proof of concept, pilot or a prototype, the tough part is recognising it as such and to be willing to throw it out.

In the enterprise world this is really difficult because managers hate seeing those hours produce seemingly nothing, when in fact the product that follows will inevitably be orders of magnitude better in all respects than the throw-away that preceded it. Personally, the single biggest payoff I've gained from this approach is a cleaner architecture that cleanly separates concerns, allowing individual components to be refined even further.

Does it though? I think he might be getting to the core of something important: that trashing the entire thing leaves in your brain the essence, that important thing that we drive so hard to make simple and obvious. Maybe this whole process is a hack to getting at that, even if it seems chaotic and stressful.

Your skepticism is very warranted. I was skeptical at first too, but we've used this method a few times now. It seems to help everyone to see it, then build it.