All writing

Build Logs · Subscribers · 13 Sep 2026 · 2 min

Stop paying frontier prices for formatting

My agent system routed work badly in both directions: expensive models doing trivial jobs, cheap models failing hard ones. How I think about matching the model to the job.

When I audited my agent system, the model setup was wrong in two directions at once.

Some jobs leaned on cheap, fast models for work that needed judgement, and the output showed it. Other jobs used a far more capable model than the task needed. And some of the model config was simply stale: old versions, wrong pricing metadata, so any cost-aware routing was working from bad numbers.

The fix is not “use the best model” or “use the cheapest model”. It is match the model to the job, and decide that on purpose.

Newsletter · weekly

The Build Log

One system teardown a week. What I built, what it cost, what broke, and the exact prompts that did the heavy lifting. No tool listicles.

  • A full build teardown with architecture
  • The prompt or config that unlocked it
  • The failures, with root causes

Weekly. Unsubscribe in one click. Read past issues.