Recently, while developing Agents with Microsoft Foundry, I noticed a confusing area: Microsoft has a series of "IQ" concepts — Work IQ, Fabric IQ, Foundry IQ, and Web IQ. They all seem to "provide context for Agents," but the context they supplement is actually different.
This points to a core problem in enterprise Agent engineering: context engineering. If prompt engineering is about "how to ask the model," then context engineering is about "what context to give the model." Microsoft IQ can be seen as Microsoft's enterprise-level implementation in this direction: packaging the scattered context inside and outside the organization into four consumable intelligence layers for Agents.
Here's the summary: Microsoft IQ consists of four capabilities:
- Work IQ: Understanding how employees work.
- Fabric IQ: Understanding business data and business entities.
- Foundry IQ: Understanding enterprise knowledge, policies, and authoritative documents.
- Web IQ: Understanding fresh information from the external world.
What Problem Does Each Layer Solve?
Enterprise Agent context typically comes from four sources, and each Microsoft IQ layer corresponds to one:
| Context Source | Corresponding IQ | Core Question | Typical Representation |
|---|---|---|---|
| Human activities and relationships | Work IQ | Who, when, with whom, did what | Microsoft Graph, collaboration signals |
| Business facts and metrics | Fabric IQ | What is the current state of the business | Semantic model, ontology, Graph |
| Enterprise knowledge and rules | Foundry IQ | What do authoritative sources in the organization say | Knowledge base, agentic retrieval |
| External real-time information | Web IQ | What is happening in the external world | Web search, fresh content |
Work IQ: Understanding "How People Work"
Work IQ is the workplace intelligence layer on the Microsoft 365 side. It understands work context, relationships, and patterns, enabling Agents to answer questions faster and more accurately than simple connectors.
From a context engineering perspective, Work IQ's core is activity signals centered on "people / relationships / time." It answers not "what is written in the document," but "who, when, with whom, did what."
For example: What recent email exchanges have there been with a customer? Which documents should I review before tomorrow's meeting? What has the team discussed around a particular project in Teams?
Work IQ leverages Microsoft 365 tenant data, including SharePoint, OneDrive, Outlook, Teams, and collaboration signals in Microsoft Graph, while respecting Microsoft 365 permissions and organizational boundaries. It is particularly suitable for employee productivity Agents: meeting preparation, email summarization, project context梳理, etc.
Fabric IQ: Understanding "What Is the Current State of the Business"
Fabric IQ is the IQ workload within Microsoft Fabric. Official documentation states: "Fabric IQ provides context on the state of your business." It is not simply about letting Agents query tables, but about elevating data to the business language layer, enabling both humans and Agents to understand data around business concepts, goals, and rules.
From a context engineering perspective, Fabric IQ solves the problem of "how business data is correctly understood by Agents." Enterprise data originally resides in tables, fields, and schemas, but business cares about: Who is the Customer, what is the relationship between Shipment and Order, how is a certain KPI calculated, and what actions should be triggered when anomalies occur.
Fabric IQ organizes these business concepts, relationships, rules, and actions through OneLake, Power BI semantic models, ontology, Graph, data agent, and operations agent. It is suitable for analytical and operational Agents: business analysis, supply chain anomaly detection, metric interpretation, and real-time operational response.
Foundry IQ: Understanding "Where Enterprise Knowledge Lives"
Foundry IQ leans more toward the managed knowledge layer for Agents. Agents need context from scattered enterprise content, but models have knowledge cutoffs and cannot access enterprise private data on their own. Foundry IQ can create configurable, multi-source knowledge bases, enabling Agents to provide permission-aware responses based on organizational data.
From a context engineering perspective, this is close to RAG, but not a hand-written simple vector library. Foundry IQ's knowledge base can connect to sources such as Azure Blob Storage, SharePoint, OneLake, and public web data, and use agentic retrieval (query rewriting, multi-round retrieval, source ranking, citation generation, etc.) to return grounded answers with citations.
It solves the problem of "how Agents reliably access enterprise knowledge:" Where are company policies? How is product documentation explained? What related content is in historical solutions? Can the answer provide citations?
It is suitable for knowledge Q&A, policy assistants, product documentation assistants, and pre-sales solution assistants.
Easy to Confuse: Both Foundry IQ and Fabric IQ Mention OneLake — What's the Difference?
- Fabric IQ handles data in OneLake that has already been modeled as business semantics (tables, metrics, Graph, semantic models). It enables Agents to understand "business facts."
- Foundry IQ handles unstructured content in OneLake such as documents, templates, and solutions that serve as enterprise knowledge sources. It enables Agents to understand "enterprise language / rules."
One is the "business data layer," the other is the "enterprise knowledge layer."
Web IQ: Understanding "What Is Happening in the External World"
Web IQ is the fourth capability. Official Microsoft IQ documentation states that "Web IQ provides AI systems and agents with fresh, real-world intelligence from across the web."
From a context engineering perspective, it solves the problem of "externally changing information:" What is the latest industry news? What have competitors released? What is the impact of new policies, regulations, or vulnerabilities?
If the first three are more about finding context within the organization, Web IQ supplements changes from outside the organization.
Easy to Confuse: Foundry IQ Can Also Connect to Public Web Data — Isn't That Redundant with Web IQ?
No, it's not redundant.
- Web data in Foundry IQ is externally curated by the enterprise, such as competitor whitepapers, regulatory pages, and industry reports that have been crawled into the knowledge base. It has already been incorporated into the enterprise knowledge boundary.
- Web IQ is real-time, open, general web intelligence that does not require pre-indexing, supplementing "what is happening in the external world right now."
One is "external materials managed by the enterprise," the other is "the real-time external world."
The Differences Among the Four
| IQ | Focus | Typical Question | Typical Data | Agent Function |
|---|---|---|---|---|
| Work IQ | How employees work | Where did I leave off with the client yesterday? | Emails, meetings, Teams, files, personnel relationships | Understand work context |
| Fabric IQ | Current state of business | What is this customer's renewal risk? | OneLake, semantic models, Ontology, Graph, real-time/analytical data | Understand business concepts and metrics |
| Foundry IQ | Where enterprise knowledge lives | What does our service agreement say about this type of issue? | Documents, knowledge bases, SharePoint, Blob, OneLake, enterprise-curated web materials | Obtain citable authoritative knowledge |
| Web IQ | What is happening in the external world | Has the customer been affected by recent industry negative news? | Web information, news, public pages, real-time external materials | Obtain fresh external context |
A Context Engineering Example
Suppose we are building a customer meeting preparation Agent:
Help me prepare for tomorrow's meeting with Contoso.
This is not a simple prompt, but an orchestration of multi-layer context:
- Use Work IQ to pull recent emails, meetings, and Teams discussions related to Contoso, establishing "people" and "timeline" context.
- Use Fabric IQ to query Contoso's business status, such as ARR, orders, risks, renewals, and service ticket trends.
- Use Foundry IQ to query product documentation, customer solutions, contract policies, and delivery templates.
- If needed, use Web IQ to supplement recent public news, industry trends, or competitive information about Contoso.
- Finally, the Agent compiles a briefing, risk points, and suggested questions.
The key is not "which IQ is stronger," but that they supplement different contexts. For example, if you only connect Foundry IQ to query documents without knowing that the customer recently expressed strong dissatisfaction in Teams, the Agent's response is likely to be "correct but contextually detached."
Summary
The real difficulty with Agents is not just whether the model can reason, but whether it has the right context. Microsoft breaks this context into different layers:
- Work IQ enables Agents to understand work like a colleague.
- Fabric IQ enables Agents to understand data like a business expert.
- Foundry IQ enables Agents to find authoritative materials like a knowledge manager.
- Web IQ enables Agents to monitor external changes like a researcher.
From a context engineering perspective, the core value of Microsoft IQ lies in: packaging the scattered, heterogeneous, and permission-complex context within the enterprise into standardized capabilities that Agents can directly consume. You don't need to write Graph connectors, build semantic layers, maintain vector libraries, or integrate web searches yourself — you simply choose the appropriate IQ layer for the scenario.
If an enterprise Agent only connects to the model, it easily becomes an assistant that "speaks well but doesn't understand the organization." When it can simultaneously understand work, business, knowledge, and external changes, it comes closer to a truly deployable enterprise Agent.
Top comments (0)