Ideas

Early-stage concepts. Each earns research, experiment, prototype, or product status — or gets retired in public.

Economic model routing

Route each task to the cheapest model measured capable of it, using observed cost-per-task and success rates per model — not benchmarks, not vibes. Status: IDEA.

PR-level AI cost accounting

Extend per-task token costs to fully-loaded cost per merged pull request: model spend plus review and rework time. If it can’t be measured per PR, it can’t be managed. Status: IDEA.

Provider cost comparison as buying leverage

Continuously measured per-model costs across providers turn pricing pages into negotiable numbers. Know exactly what each provider costs you per task before renewing anything. Status: IDEA.