Microsoft's 4-IQ Framework: The Enterprise Knowledge Layer You Need to Understand
At Microsoft Build 2026, Microsoft announced Web IQ — real-time internet grounding for Foundry agents. But Web IQ is one piece of a larger architectural framework Microsoft is calling the IQ family. Understanding all four IQs and how they compose together is now essential knowledge for any Azure architect building enterprise AI agents.
What the IQ Framework Actually Is
The IQ family is not a product — it is a knowledge grounding layer for AI agents. Each IQ gives your Foundry agent access to a different category of information. Agents that use IQ grounding produce responses backed by real organisational, business, structured, or internet data rather than relying solely on model weights.
Think of it this way: the AI model handles reasoning; the IQ layer handles what the agent knows about right now.
Work IQ — Organisational Signals from Microsoft 365
Work IQ reached general availability on 16 June 2026. It gives agents access to organisational context from the Microsoft 365 ecosystem: email signals, calendar patterns, document history, meeting context, and people relationships.
This is powerful for HR agents, project management agents, and executive assistants. An agent with Work IQ knows who owns a decision, who last worked on a document, and what conversations have already happened — without anyone having to brief it.
Fabric IQ — Structured Business Data with an Ontology Layer
Fabric IQ connects agents to your Microsoft Fabric data estate: lakehouses, warehouses, and semantic models. The key differentiator is the ontology layer — a structured understanding of how your business data entities relate to each other.
Where a raw SQL query returns rows, Fabric IQ returns business-meaningful context. An agent asking about sales performance gets not just numbers but the relationships between regions, products, teams, and time periods — semantically grounded.
Azure IQ — The Foundry Orchestrator
Azure IQ (sometimes referred to as Foundry IQ) is the orchestration layer that calls all other IQs. It handles routing — deciding which knowledge sources are relevant to a given query and composing the responses from multiple IQs into a unified grounding context for the agent.
This is the IQ your orchestrating agent uses. It does not hold data itself; it knows how to get it from the right source at the right time.
Web IQ — Real-Time Internet Grounding
Web IQ is the newest addition and the one announced at Build 2026. Unlike a standard web search that returns links, Web IQ returns passage-level evidence objects — structured excerpts from web sources with attribution metadata. Foundry agents consume these as grounded context, not as raw search results to parse.
This distinction matters for hallucination reduction. The agent is not reasoning about search result snippets — it is working with structured evidence objects that include source, credibility signals, and relevance scores.
How the Four IQs Compose
The real architectural power comes when the IQs work together. Consider a procurement agent handling a supplier risk query: Azure IQ orchestrates the call; Work IQ retrieves internal communication history with the supplier; Fabric IQ queries supplier performance data from your data warehouse; Web IQ pulls recent news and public financial signals. The agent reasons across all four knowledge sources in a single response.
What Architects Should Do Now
- Audit your agent knowledge requirements before selecting IQ sources — not every agent needs all four
- Map your Fabric semantic models now so Fabric IQ grounding is meaningful rather than raw table access
- Treat Web IQ evidence objects as citations — surface attribution to end users for transparency
- Design Azure IQ routing logic to control cost — IQ calls are not free, and promiscuous grounding will inflate token usage
The 4-IQ framework is Microsoft's answer to the enterprise knowledge grounding problem that has held back production AI agent deployments. Architects who understand the composition model will build significantly more capable and trustworthy agents than those treating each IQ in isolation.
