> Doherty Threshold: productivity soars when a computer and its users interact at a rate (<400ms) that means neither has to wait for the other

This is why I strongly prefer smaller models for programming.[0] They're fast enough that the activity stays real-time.

It also forces you do to split the work into smaller chunks and verify it continuously. So you stay active and engaged, and your mental model never gets out of sync.

---

[0] I once gave three simple code changes to a big model and a small model. They both completed the tasks successfully. The big model took 3 times longer and cost 10 times as much.

In that moment I switched my definition of Best Model from "tops the benchies" to "the smallest, fastest, cheapest one that can reliably do the actual job."