Microsoft Fabric becomes the training ground for enterprise AI agents to understand business operations

At FabCon and SQLCon 2026 in Barcelona, Microsoft demonstrated how its Fabric platform now serves as a comprehensive context layer for AI agents operating across enterprises. The core challenge, according to Fabric's leadership, is deceptively simple: autonomous agents have no persistent memory, requiring explicit instruction about the company they're working for, the data they can access, and the business rules governing decision-making.

The Context Problem#

AI agents face a fundamental limitation—each interaction begins without institutional knowledge. As Fabric CTO Amir Netz explained, agents must be told where they are before they can work: "You are now working for Microsoft. You are now working for Wells Fargo. You are now working for Emirates."

Enterprises deploying agents are learning this challenge firsthand. Companies unify their data and run AI models against it, only to discover the results don't match expectations. This mismatch reveals a missing ingredient: the structured business context that gives agents accurate decision-making capability.

Yitzhak Kesselman, corporate vice president overseeing Fabric IQ, notes that advanced customers build evaluation frameworks for their agents and quickly recognize the need to encode business context. Enterprise leaders increasingly demand proof of agent value through measurable before-and-after outcomes.

Four Layers of Intelligence#

Microsoft is assembling what it calls Microsoft IQ—a suite of context systems for different information types:

• Work IQ covers email, Teams, and SharePoint communication
• Foundry IQ handles documents and manuals
• Web IQ accesses public internet data
• Fabric IQ focuses on operational business state, combining unified data from OneLake, trusted metrics from Power BI semantic models, and business rules through ontologies and real-time data streams

Fabric IQ now feeds Microsoft 365 Copilot by default and is accessible to external agents via Model Context Protocol (MCP), allowing third-party systems to query business context without copying data.

Unifying Data at Scale#

OneLake, Fabric's central storage layer, anchors the system by consolidating data from first-party and third-party sources. This week, Microsoft announced new integrations including two-way synchronization with Salesforce Data Cloud 360, general availability of BigQuery mirroring, ClickHouse workload support, and dbt's Fusion engine arriving in Fabric within weeks.