Power BI used to be a tool you clicked your way through. Then PBIP turned your semantic models and reports into plain text (TMDL, JSON), and suddenly an entirely different toolbox showed up at the door: AI agents that can read, write, and reason about your model the same way they reason about code.

This session is about what happens once you let those agents in. Not the marketing kind of "AI in Power BI", but the practical, sometimes scrappy, often surprisingly powerful kind: GitHub Copilot in agent mode, working on your PBIP files, with the right context loaded, the right skills wired up, and the right MCPs connecting it to Fabric.

We'll start with a quick tour of why PBIP matters and what an "agent" actually is in this context. From there we go hands-on. Expect a lot of demos: Copilot writing DAX in TMDL, generating documentation, doing bulk model edits, reviewing pull requests, and chaining all of it together into workflows that feel a lot less like clicking and a lot more like delegating. We'll wire up an MCP so Copilot can talk to Power BI Desktop directly, and we'll show what changes when your agent stops guessing and starts looking things up.

Throughout, we'll keep one question front and center: where does the agent earn its keep, and where is it still happily wrong with confidence? You'll leave with a clear sense of what's real today and a workflow you can replicate on Monday morning.