Why Your AI Agent Is So Expensive and How to Fix It.

AI coding agents are powerful, but they can burn through tokens fast. Two people can use the same tool for the same task and end up with very different bills or usage limits. Why?

The reason is usually not the model. It is what we send to it. An agent works in steps, and at every step it can send the model a large amount of context again: instructions, tool descriptions, earlier messages and file contents. Small habits and setup choices add up to big differences in cost.

This talk explains where tokens actually go in an AI agent, and simple practical ways to spend fewer of them without getting worse results. No prior experience with AI agents is needed.

The following will be covered:

• What tokens and the context window are, in simple terms

• Why agents use so many tokens: how every step re-sends context

• The hidden costs: long conversations, large files and tool descriptions

• Practical ways to reduce usage: focused context, clear instructions, caching, and choosing the right model for the task

• How to give agents tools efficiently (MCP, skills, command-line tools) and what each cost

• A short exercise: find the waste in a real-life agent workflow

1. Where tokens go when an AI agent works, and why costs add up.

2. A simple checklist of habits that reduce token use.

3. How context, tools and model choice affect cost and quality.

4. How to spot waste in your own agent workflow.