For a bootstrapped business with no runway to burn, an uncontrolled AI bill is an existential threat. Margin is the whole game, and inference costs erode it silently.

The cost-control framework

  1. Right-size the model. Use the smallest model that clears the quality bar for each task. Frontier models for trivial jobs is money on fire.
  2. Cache aggressively. Repeated or similar requests should never hit the model twice.
  3. Cap per-account usage. Tie AI allowances to plan tier so one heavy user can't sink the margin.
  4. Meter and alert. Watch cost-per-account daily; a runaway integration should page you, not surprise you at month-end.

Bottom line: for a bootstrapper, AI cost discipline is survival — smallest model that works, cache everything, and cap by tier.