Building in Fabric got radically easier. Fabric Copilot generates whole workloads from a prompt. Modern Report Builder puts agentic reporting in Power BI Desktop. Developers are vibe coding with GitHub Copilot, and your users are chatting with their data in Copilot on models never prepared for AI.
I love all of it; I'm all about adoption. But each one is a new governance problem. Who can generate workloads, and which items? When a report builds itself, where did that number come from? How do we stop legacy semantic models polluting Copilot answers?
I'm relaunching the Power BI governance framework that built my career as a Fabric framework for the age of vibe coding: the culture, the processes that ground it, and the technology patterns that enable it. Governance pros need to get comfortable with noise: accessible sandboxes, robust deployment routes, clear endorsement policies.
We're past wondering if people will build with AI. The job now is knowing what they've built.