THE ECONOMICS OF AGENTS

CHAPTER SUMMARY Agents spend money mainly through tokens, and input and output tokens are priced separately. Because agents loop and carry a growing history, they use far more tokens than simple chatbots. Model inference is usually the largest source of both cost and delay, so optimize model calls before tool calls. Total operating cost includes sandboxes, observability, storage, evaluation, human time, and failures, not only the token bill. FinOps for AI is the routine of measuring, attributing, budgeting, reviewing, and optimizing spend. KEY TERMS REVIEW Token: a small piece of text that a model reads or writes, and the unit providers bill by. Input tokens: tokens sent to the model. Output tokens: tokens the model generates. Inference: running a trained model to get an answer. Latency: delay between a request and its response. Sandbox: an isolated environment for safely running code. Observability: the ability to see what a system did through logs, traces, and metrics. Total operating cost: everything required to run the agent, not only token charges. FinOps: disciplined financial management of technology spending.