Can a software factory intelligently choose a model for each dispatch while taking cost into account?
Build dispatch policies that match task complexity to the least expensive model that can reliably complete the work.
Starting point
The economics of CLI harnesses such as Claude Code and Codex depend on subsidies that may not exist forever. The factory should pencil out at API rates while still taking advantage of fixed-cost developer plans when they are available.
What to track
- The optimal mix of coder plans and direct API usage.
- How task complexity predicts model success, retries, and review cost.
- Whether a usage cycle should end with nearly all included inference capacity consumed.
- Which durable work is worth doing with otherwise unused capacity.
Living record
Dispatch-policy changes, model evaluations, and realized cost data belong here.