Fabric IQ, Work IQ, and Foundry IQ are three separate Microsoft context sources for AI agents, and an agent sees only the ones your team connects to it. Fabric IQ covers business entities over OneLake, Work IQ covers how people work across Microsoft 365 and business apps, and Foundry IQ covers curated documents.
The Microsoft IQ documentation hub (Microsoft Learn, 2026) lists the three with Web IQ as “four interconnected capabilities,” side by side but never compared. Choosing among them means reading each product’s docs for scope, status, identity, and billing.
| Dimension | Fabric IQ | Work IQ | Foundry IQ |
|---|---|---|---|
| Grounds agents in | Business entities and KPIs over OneLake | Mail, meetings, chats, files, Dynamics 365, Power Platform | Policies and authoritative documents |
| Status (September 2026) | GA announced at Build 2026; Learn still labels workload and ontology preview | Generally available since June 16, 2026 | GA or preview by Search Service REST API version |
| Core items | Ontology, semantic model, graph, planning, agents | Chat, context, tools, workspaces | Knowledge bases, knowledge sources, agentic retrieval |
| Agent interface | Per-item MCP endpoints; a Foundry tool | A2A, remote MCP server, REST | Knowledge base REST API, Azure SDKs |
How Fabric IQ, Work IQ, and Foundry IQ split organizational context
Fabric IQ, Work IQ, and Foundry IQ differ by which part of the organization they describe: business data, the way people work, and curated documents. The Microsoft IQ hub (Microsoft Learn, 2026) gives each a one-line role:
- Work IQ “provides live understanding of how employees work, with context on people, collaboration, and workflows.”
- Fabric IQ “provides the live state of the business, so agents understand business entities and their relationships, properties, actions, and rules.”
- Foundry IQ “provides curated institutional knowledge on the context of policies, authoritative documents, and reusable knowledge bases.”
The Foundry IQ overview (Microsoft Learn, 2026) states the relationship: “Each IQ workload is standalone, but you can use them together to provide comprehensive organizational context for agents.”
Confusion starts at the edges. Foundry IQ lists OneLake as a knowledge source, so the same files can reach an agent through two IQs: one returns entities and relationships, the other cited passages, the split between a context graph and a knowledge graph. James Serra, a Microsoft solution architect, wrote on his personal blog (James Serra, 2026): “I was confused too.”
Pick by the slice an agent must see. Each IQ answers only for its own, so the enterprise context layer an agent runs on is whatever set of slices a team deliberately assembled.
What is Fabric IQ?
Fabric IQ is the Microsoft Fabric workload that models business concepts such as Customer or Shipment over OneLake data, so people and agents reason in business terms. According to Microsoft’s Fabric IQ overview (Microsoft Learn, 2026), its core items are the ontology (preview) and the Power BI semantic model, from which ontologies can be generated.
Microsoft “announced the general availability of Fabric IQ” at Build 2026, per the Microsoft Fabric blog (Microsoft, 2026), yet Learn still labels the workload and ontology preview. Connecting a Foundry agent to it comes “without a service-level agreement,” per the Foundry how-to (Microsoft Learn, 2026).
Access rides on Fabric’s own model (Microsoft Learn, 2026): requests “honor Fabric permissions and governance policies,” and every invoking user needs a Fabric license. Keith Kirkpatrick, VP and Research Director at Futurum Group, and co-authors wrote that Fabric IQ “connects business data to how the business is described” (Futurum Group, 2025).
Core components of Fabric IQ
- Ontology (preview) defines entities, relationships, rules, and actions, queried through NL2Ontology; see ontology vs. semantic layer.
- Power BI semantic model carries curated measures and hierarchies.
- Graph traverses connected data, GA since June 2026 per the What’s New archive (Microsoft Learn, 2026).
- Data and operations agents answer domain questions and act on live data.
Fabric IQ is the one of the three built to model structured business meaning, the ground an ontology 101 explainer covers. Its view stops at OneLake: the agent knows what a Shipment is, not the Teams thread where it was escalated.
What is Work IQ?
Microsoft calls Work IQ a workplace intelligence layer that lets agents “access and reason over organizational and business data, context, and tools,” per the Work IQ overview (Microsoft Learn, 2026). Beyond Microsoft 365 mail, meetings, chats, and files, it includes data from Dynamics 365 and Power Platform.
The Work IQ CLI documentation (Microsoft Learn, 2026) dates general availability to June 16, 2026; support for preview builds such as 0.4.x and 0.5.x “is no longer offered.”
According to Microsoft Learn (2026), the Work IQ MCP “collapses hundreds of operations into just 10 generic tools” across mail, calendar, files, people, chat, and sites, a verbs-plus-resource-paths design an MCP server implementation guide walks through.
Authorization pairs a small set of broad permissions with a Rego-based policy engine that checks path, method, identity, and content on every request. Actions are user scoped, and billing is usage-based, “independent of Microsoft 365 Copilot licensing” (Microsoft Learn, 2026).
Core components of Work IQ
- Chat uses A2A for agent-to-agent delegation and REST for web apps.
- Context grounds signals internally, so agents skip building retrieval pipelines.
- Tools is that 10-tool MCP surface of verbs.
- Workspaces gives long-running workflows SharePoint Embedded storage.
Work IQ sees how the organization operates, through one user’s permissions, and holds no model of what a revenue metric means. An agent that needs both needs both sources, the case in why AI agents need an enterprise context layer.
What is Foundry IQ?
Foundry IQ is a managed knowledge layer that turns enterprise content into “permission-aware knowledge bases” for agents, per Microsoft’s Foundry IQ overview (Microsoft Learn, 2026). Sources include Blob Storage, SharePoint, OneLake, and the web, with Azure AI Search underneath.
Its status note (Microsoft Learn, 2026) is set feature by feature: “Some Foundry IQ features are now generally available, while others remain in preview. Availability depends on the Search Service REST API version you use.”
Permissions are documented and conditional. The overview’s capability list (Microsoft Learn, 2026) says Foundry IQ can sync access control lists for supported sources, “honor Microsoft Purview sensitivity labels,” and run queries under the caller’s Entra identity. The Foundry IQ FAQ (Microsoft Learn, 2026) adds a caveat: document-level controls apply only where a source supports them and sync is configured; remote SharePoint sources enforce them out of the box.
Billing follows Azure AI Search tokens after a free allocation, with LLM query planning billed separately. The FAQ (Microsoft Learn, 2026) says one knowledge base serving many sources sets it apart from classic retrieval, the line context engineering vs. RAG draws.
Core components of Foundry IQ
- Knowledge base picks sources and sets retrieval reasoning effort.
- Knowledge sources are indexed (chunked and vectorized) or remote (queried on demand).
- Agentic retrieval splits a question into parallel subqueries and returns citations.
Foundry IQ is the one built to answer from documents. Because its permissions vary by source, “which sources, which labels, which identity” gets settled per knowledge base before an agent reads from it.
Fabric IQ vs. Work IQ vs. Foundry IQ: head-to-head comparison
Microsoft’s three IQs diverge on identity, billing, and permissions, and share one purpose: grounding an agent in context the model cannot see alone.
| Dimension | Fabric IQ | Work IQ | Foundry IQ |
|---|---|---|---|
| Scope | OneLake, including shortcuts | Microsoft 365, Dynamics 365, Power Platform | Admin-configured sources |
| Identity | Delegated user auth via Entra app or managed OAuth | Broad permissions plus per-request Rego policy | Caller’s Entra identity |
| Billing | Fabric license; F2+ for data agents | Usage-based meter | Azure AI Search tokens, plus Azure OpenAI |
| Permission surface | Fabric permissions and policies | Rego engine; tool calls logged | ACL sync, Purview labels where supported |
| Primary consumer | Fabric, Foundry, Copilot Studio agents | Microsoft 365 Copilot, custom agents | Foundry agents, custom apps |
| How it’s queried | NL2Ontology, semantic models, MCP | 10 MCP tools, A2A, REST | Keyword, vector, hybrid, agentic |
| Blind spot when wired alone | Conversations and policy documents | Governed metric definitions | Live business state |
Picture a supply-chain team shipping a Foundry agent grounded in Fabric IQ’s ontology. Asked which shipments breached cold-chain thresholds, it answers well by traversing Order to Shipment to Temperature Sensor, the example in the Fabric IQ overview (Microsoft Learn, 2026). Asked what the carrier heard on Tuesday’s Teams call, or what the returns policy says, it has nothing, and no error says why.
A Work IQ-only agent can summarize that call but not traverse the ontology’s shipment-to-sensor chain; a Foundry IQ-only agent can quote the returns policy but not see live shipment state.
Each IQ boundary is also a context boundary, which is how multi-agent memory silos form inside one vendor’s stack, and why a memory layer and a context layer solve different problems.
How do Fabric IQ, Work IQ, and Foundry IQ work together?
Fabric IQ, Work IQ, and Foundry IQ combine only when a team connects each one to the agent. The Foundry IQ FAQ (Microsoft Learn, 2026) adds that Copilot and Foundry IQ knowledge sources “aren’t interoperable”: one built for either can’t be reused in the other.
The Microsoft IQ hub links separate guides for Fabric IQ and Work IQ in Foundry, so composition happens in the AI agent harness: one tool, one identity grant, one meter per IQ.
At FabCon Europe 2026, Hugo Barona’s session T04, “Fabric IQ + Foundry: The Semantic Backbone of Enterprise AI Agents,” starts from the premise that “AI agents fail when business definitions, metrics, and context are interpreted differently across systems” (ESPC, 2026); W12, W46, and TH14 appear in the FabCon Europe 2026 sessions by role.
Fabric-only agent
An agent that only answers questions about Fabric data may never feel the gap. One identity model and one bill is a sound design for that scope, and the usual start for a single-agent system.
Cross-stack agent wiring all three
An agent that needs collaboration signals, Fabric data, and policy documents must be wired to all three: broad context, but three grants, meters, and permission models to keep aligned, the load AI agent identity covers.
Microsoft 365 plus Fabric (partial wiring)
A Dynamics 365-heavy team might connect Work IQ and Fabric IQ and skip Foundry IQ. Daily questions land; contract and policy questions miss, the pattern behind enterprise context silos.
When to prioritize one over the other
- Fabric IQ first when agents live inside Fabric, OneLake, and Power BI.
- Work IQ first when the priority is Copilot, Microsoft 365, or Dynamics 365 grounding.
- Foundry IQ first for retrieval over curated documents, often for Azure AI Foundry agents.
- Several at once when one agent’s questions cross data, collaboration, and documents; context management in multi-agent systems gets harder with each.
Where a context layer fits alongside Microsoft’s IQ stack
Each IQ is bounded to one part of the Microsoft stack. An organization also running Snowflake, Databricks, or Salesforce keeps definitions and lineage outside all three, so the core components of a context layer must reach those too, raising who owns the context layer early.
Atlan is the Context Layer for AI, built to sit across a mixed estate, Power BI included. Its Context Agents read lineage, SQL, human annotations, usage patterns, BI semantic logic, and agent traces. Orion maps what a term means everywhere; Sage finds where two teams define a metric differently and locks in one answer, Fabric IQ’s ontology work made cross-platform on a context graph.
For Fabric-committed teams the scopes coexist. Ask the questions in Fabric AI agent governance questions; the build order is in how to implement an enterprise context layer for AI, a context engineering problem at heart.
Which IQ an agent needs depends on the question it has to answer
Microsoft frames standalone workloads as modularity, which holds for a single-platform agent. Friction arrives once questions cross platforms and each added IQ brings another grant, bill, and label to reconcile, the semantic layer problem in a new form.
Treat “which IQ” as a per-agent wiring decision, and date-stamp each piece’s status. As IQ-branded surfaces multiply, test each new source against an AI-ready data checklist and what makes data AI-ready before the broader AI-ready data audit.
FAQs about Fabric IQ vs. Work IQ vs. Foundry IQ
1. Is Fabric IQ generally available or still in preview?
Both. Microsoft announced general availability at Build 2026, but as of September 2026 Microsoft Learn still labels the IQ workload, the ontology, and the Foundry agent connection preview.
2. Does an agent connected to Fabric IQ automatically see Work IQ or Foundry IQ context?
No. Microsoft describes each IQ workload as standalone. An agent grounded in Fabric IQ sees nothing from Work IQ or Foundry IQ until a team adds each as a separate tool, with its own identity and billing.
3. What is the difference between Fabric IQ and Foundry IQ?
Fabric IQ models business entities, relationships, and KPIs over OneLake data through an ontology and semantic models. Foundry IQ retrieves cited passages from admin-connected knowledge sources through Azure AI Search. Both can reach OneLake, with different results.
4. Is Microsoft IQ a separate product or a feature?
Neither. Microsoft IQ is the umbrella name for four interconnected capabilities, Work IQ, Fabric IQ, Foundry IQ, and Web IQ, each documented and set up separately.
5. How does Work IQ access my private Teams and Microsoft 365 data?
Work IQ acts for a specific user and reaches only what that user may see. A Rego-based policy engine evaluates every request, each tool call is logged, and the CLI needs administrative consent in your Microsoft Entra tenant.
6. Do I need a Fabric license to use Fabric IQ?
Yes. Microsoft requires a Fabric license covering the items an agent queries, and invoking users need one too. A Fabric data agent needs F2 or higher capacity, or Power BI Premium P1 or higher.
7. Can AI agents use Web IQ and Foundry IQ at the same time?
Yes. Microsoft lists them as interconnected capabilities, and a Foundry IQ knowledge base can also pull public web data through Grounding with Bing.
Sources
- What is Fabric IQ?, Microsoft Learn (Microsoft, 2026)
- What is Foundry IQ?, Microsoft Learn (Microsoft, 2026)
- Foundry IQ FAQ, Microsoft Learn (Microsoft, 2026)
- Connect agents to Microsoft Fabric with Fabric IQ (preview), Microsoft Learn (Microsoft, 2026)
- Work IQ overview, Microsoft Learn (Microsoft, 2026)
- Microsoft Work IQ CLI, Microsoft Learn (Microsoft, 2026)
- Microsoft IQ documentation, Microsoft Learn (Microsoft, 2026)
- What’s new in Microsoft Fabric archive, Microsoft Learn (Microsoft, 2026)
- FabCon Europe 2026: The sessions we’re most excited to bring to Barcelona, Microsoft Fabric Blog (Microsoft, 2026)
- Fabric IQ + Foundry: The Semantic Backbone of Enterprise AI Agents, ESPC FabCon Europe 2026 programme (ESPC, 2026)
- Making Sense of Microsoft’s AI Strategy: Work IQ, Fabric IQ, Foundry IQ, James Serra (James Serra, 2026)
- Microsoft Ignite 2025: AI Agent 365, Anthropic on Azure, Security Advances, Futurum Group (Futurum Group, 2025)
