Microsoft Fabric is where AI agents learn how the business works

Microsoft is pushing Fabric beyond being a data platform and into something closer to an operating manual for AI agents. At FabCon and SQLCon in Barcelona, the company laid out a stack meant to give agents the context they usually lack: what happened before, what is happening now, and what a business is supposed to do next.

The centerpiece is Fabric IQ, the context layer inside Fabric. Microsoft says it now feeds Microsoft 365 Copilot by default, can be queried by outside agents over MCP, and will soon help Power BI turn the same business definitions into apps. Ontologies, which add business rules on top of the data, are also moving further into preview. Amir Netz, Fabric CTO, compared agents to someone who forgets everything each time they open their eyes. The fix, in his telling, is simple: tell them where they are.

That context starts with data, and Microsoft keeps hammering on that point. Yitzhak Kesselman, who runs Fabric IQ, said customers often discover only after the fact that they need context for their agents. Microsoft’s bet is that OneLake can be the gathering point, pulling in data from the services companies already use. Shortcuts and mirroring into OneLake are free, and Netz said storage under management there is growing 300% year over year.

Microsoft also used the week to add more pipes into that system. It announced a two-way integration with Salesforce Data Cloud 360, general availability for mirroring from Google BigQuery, a ClickHouse workload that runs the engine directly against OneLake, and dbt’s Fusion engine coming in the next couple of weeks. Security is part of the pitch too: mirrored security roles are headed to public preview, carrying permissions along with the data rather than dropping them on the floor.

The larger message is that Microsoft wants Fabric to cover every time frame a business cares about. Semantic models cover the past. Real-time intelligence covers the present. Fabric Planning, which went generally available in July, covers budgets, forecasts, and targets. And ontologies are there to make sure an agent knows that a pilot is not just a row in a table, but someone with rules, certifications, and rest hours that matter. Microsoft says the answers are now generally available, while the actions are still filling out the preview list. That feels about right for enterprise AI: the context comes first, and the magic trick follows later.