Joshua Vaz
For founders · checklist

Where the AI bill actually goes

Nobody sends you an itemised AI bill. This is what would be on it if they did, and which of the eight lines you can actually move.

What is on it

  • Input tokens: everything you send, including the instructions
  • Output tokens: usually two to five times the price of input
  • Retries: every failed call you paid for twice
  • Context stuffing: the history you resend on every single turn
  • Embeddings: cheap each, expensive when you re-index everything
  • Vector storage: a monthly floor that does not go down
  • The wrapper's margin: what your vendor adds on top
  • Idle capacity: reserved throughput nobody is using

Send it to me

Put your address in and the download appears on the next screen. A copy lands in your inbox too, so it is there when you need it again.

You also get the newsletter, every second Tuesday. Unsubscribe in one click.

Preview of Where the AI bill actually goes