Pick a Model Per Subagent, Not Per Session

Custom subagents, the sub-tasks you delegate to their own context window, take a model: field, and most people never set it.

It cuts both ways. A security reviewer restricted to read-only tools can run on the strongest model, because that judgement is worth the tokens. High-volume investigation subagents can run on a cheaper, faster one, because reading forty files and summarising is not the hard part.

Reserve the expensive model for architectural decisions, tricky debugging, and long-horizon work.

The default, everything on one model, is either overpaying for grep or underpaying for judgement. Usually both, in the same session.