A Fabric Data Agent is easy to create and hard to trust. It answers almost any question with confidence, and without a way to grade those answers, you find out it's wrong when a user does.

This session follows a Data Agent through three stages. First, making it good: choosing and scoping data sources, writing instructions, and adding example queries that teach it how your data works.

Second, testing it at scale. One good answer in a chat window proves little, so we'll look at how the Fabric Data Agent SDK runs a whole set of questions against an agent, judges the answers, and shows where it fails.

Third, calling it from Microsoft Foundry. Through Fabric IQ, a Data Agent becomes a tool that a Foundry agent can call alongside its other sources. We'll cover how that works, what Foundry adds, and what to consider around routing, permissions, and monitoring.

You'll leave knowing how to build a Data Agent, prove it works, and put it to use beyond Fabric.