Microsoft's three IQs are the actual grounding story

If you’ve been reading Microsoft’s agent platform posts for the last few months and squinting at the part labeled “context,” the picture is clearer this week. A useful summary lives in a Neowin write-up of the IQ family that lays out what each of the three “IQs” actually does, and the platform pieces that sit underneath them are increasingly documented in the Foundry Build 2026 recap.

The shape, simplified.

• Work IQ is the productivity-side intelligence layer: SharePoint files, Outlook mail, Teams meetings, plus a per-user memory of preferences and recent context. APIs let developers reach the same context their copilots already use.
• Fabric IQ is the data-estate-side intelligence layer: a unified semantic model over the data in OneLake, including an ontology, a graph engine for multi-hop reasoning, and data and operations agents that act on top of it.
• Foundry IQ is the agent-side intelligence layer: a managed retrieval system that grounds Foundry agents in Work IQ, Fabric IQ, Azure data services, custom apps, and the web, behind one endpoint.

Read together, that’s not three products. It’s a layered grounding stack, with productivity context, business-data context, and external-and-custom context all reachable through one set of agent-shaped APIs.

For SI and ISV partners, the practical reads are unusually concrete.

• Pick which IQ your offering sits closest to and go deep. An ISV that builds for a particular industry’s structured data should be living in Fabric IQ (ontology, semantic model, graph). An ISV that builds productivity surfaces or vertical Copilot extensions belongs near Work IQ. An agent-platform SI building cross-system workflows is operating at the Foundry IQ seam. Trying to sit across all three at once usually produces a generic story.
• The semantic and ontology layer is the new defensible artifact. Fabric IQ in particular is asking customers to invest in shared definitions of their business entities. The partners who help land that work (industry ontologies, sample semantic models, governance templates) will quietly become indispensable in a way a model-wrapper offering cannot.
• Treat memory as policy, not a feature. The Work IQ side is explicit about per-user memory across sessions. That’s powerful and it’s a governance surface. Any agent built on Work IQ inherits that memory, and the consent, retention, and deletion behavior around it. Design for that, then sell the design.
• Foundry IQ is where the integration story lives. If you’ve been building RAG pipelines from scratch for every customer, the serverless retrieval and unified knowledge story under Foundry IQ is going to change the unit economics of those projects. The partners that retool around it first will undercut the ones still hand-rolling the same pattern in eight different ways.

Microsoft has been pushing a “system around the model” thesis since Build. The IQ family is the substrate that gives the thesis teeth. Worth reading the docs as they land and shaping the next round of partner offerings against the seams, not against the slides.