Quick answer: AI agent economics comes down to cost per completed task: tokens per step times steps per task times price per token, plus infrastructure, retries and human review. The biggest levers are choosing the right model per step, controlling how context grows across steps, caching repeated input, and deciding