Module 1 · Spending less on the same work · Lesson 1
Token economics of an agent fleet
What one case costs in tokens, step by step, why input re-reading and thinking dominate agent bills, and how to turn per-call prices into cost per completed case.
By the end of this lesson you can
- Write the cost of a model call in terms of uncached input, cache writes, cache reads and output tokens, including thinking
- Explain why an agent loop's input grows quadratically with steps, and compute it for a given step count and context growth
- Build a per-step cost model of a case and identify which steps and token classes dominate
- Compute cost per completed case, including retries, human correction and cases that fall through