THE ECONOMICS OF AGENTS

WHY AGENTS SPEND MORE THAN CHATBOTS A simple chatbot answers one question with one model call. An agent works in a loop. In each pass through the loop, the harness sends the model the whole conversation so far, including the original goal, earlier tool results, and the tool descriptions. The model replies, perhaps asking to use a tool. The harness runs the tool, adds the result to the conversation, and sends everything back to the model again. Because the history grows with each step, each new call is bigger than the one before it. Imagine a group email thread where every reply includes the whole earlier conversation pasted underneath. By message twenty, the email is enormous, even though you only typed one new line. Agents behave the same way, and this is why an agent run can use fifty thousand tokens or more where a chatbot reply uses a few hundred. A cost formula you can remember is this. Cost of one call equals input tokens times the input price, plus output tokens times the output price. Cost of one run equals the sum of the costs of all its calls. Cost per month equals cost per run times the number of runs. Each of these multiplications is a place where savings or surprises can hide.