The Joker and his goons have become notorious agents running the fair city of Gotham out of control with all their new tools, causing immense token burn rates. One particularly mischievous agent keeps hammering the same MCP server over and over, pulling massive tool schemas and flooding the streets with more context than they are designed to handle.

This session walks through a practical example of building an AI Agent using Microsoft Foundry in .NET. The agent’s goal is to act as a Sales Agent for a company by connecting to the company’s CRM through an MCP server. In this solution, tokens were being consumed at an extremely high rate and needed to be brought under control. I will show how we instrumented the agent to observe token usage, identify and trace where the tokens were being spent, and ultimately applied a few techniques to reduce the burn rate and optimize token consumption.

See how Batman and Robin restore order to Gotham by stopping the Joker’s reckless agents. See how we do the same for our AI agent by bringing runaway token usage under control. Developers will leave with clear strategies for measuring token usage, spotting expensive MCP patterns, and optimizing agent architectures in the Microsoft AI stack.