Should a cheap model run your agent's loop?

Anthropic, Tobi Lutke, and Stanford's HazyResearch all arrived at the same shape in 2026: a cheap model runs the loop, an expensive model weighs in only at hard decisions. Five questions place your setup on that shape — and draft your handoff contract.

Built from Harrison Guo's "The Advisor Strategy: Agent Architecture Is a Compute Allocation Problem" — all credit to the original post.