← Aaron Figueroa

Token-efficient agents

Drop this into any repo where Claude, Codex or another coding agent does real work. It keeps the expensive model on decisions and gives the rest to cheaper models.

Download AGENTS.md →

1. Use the cheapest model that does the job well

JobModel tier
Watching: check whether something changed; answer "unchanged" or one lineSmall (Haiku-class)
Scripted runs: a named script or command list; report the exit code and summary as-is. Never edits, reviews or decidesSmall
Well-specified builds: small fixes with a named spec, plumbing on an existing pattern, docsMid (Sonnet-class), with self-checks
Design work: new modules, cross-cutting changes, anything touching auth, persistence or cachingTop model writes the spec; mid model builds it unit by unit and asks instead of improvising
Root cause after a failure; independent reviewTop model, fresh context

Name the model on every agent you start. Don't rely on the default.

2. Make the cheap tiers the workhorses

Each unit of work names four things: the model, the exact files or commands, what "done" means, and when to stop and ask. A vague brief stops with a question. It doesn't move up a tier.

If you do a sequence twice, script it. Then a small model runs the script.

3. Don't pay to wait

4. Keep context small

5. Keep the structure flat

6. Re-verify, don't trust

7. Check the split with numbers

Keep a small ledger per build: builder model, review verdicts, defects found later. Every few builds, compare. If the cheaper model's numbers are worse for a kind of job, move that job back up a tier and write down why.