If you have looked at a Microsoft keynote slide in the last six months, you have probably seen the word “IQ” attached to half a dozen products. Work IQ, Fabric IQ, Foundry IQ, and now Web IQ. It is easy to assume these are four new SKUs you need to buy. They are not. Together they make up **Microsoft IQ**, the shared intelligence layer that grounds Copilot and every agent you build in the real context of your organization.
This post breaks down what Microsoft IQ actually is, what each of the four layers does, and why it matters if you build on the Power Platform.
## The problem Microsoft IQ is trying to solve
Most organizations can spin up an agent in an afternoon. Moving that agent from a demo into production is where things fall apart. The reason is almost always the same: the agent has no real understanding of the business.
Microsoft frames the core problems like this:
– **Agents lack true business context.** Built in silos, they reason over fragmented data and one-off retrieval pipelines, missing the relationships and meaning that actually drive decisions.
– **Enterprise signals are scattered.** Data, knowledge, and workflows live across dozens of systems, so every team rebuilds the same context from scratch.
– **Agent development is brittle.** Custom plumbing and fragile orchestration logic do not scale across scenarios.
– **Security and governance are bolted on.** Without consistent, identity-aware controls, leaders cannot trust agents enough to scale them.
Microsoft IQ is the answer to all four. Instead of every agent rebuilding connections, embeddings, routing, and permissions, you stand up a shared, continuously updated understanding of your organization once, and every agent inherits it. Build one agent and it knows your company on day one. Build the next and it reasons from the same grounded foundation.
A useful way to think about it: Microsoft Graph *exposes* your data through APIs, while Microsoft IQ *turns that data into reusable enterprise intelligence* that agents can reason over. Importantly, Microsoft IQ is not a model or an LLM. It is the context and governance layer that sits between your models and your business.
## The four layers of Microsoft IQ
Microsoft IQ is made up of four integrated layers. Each is a standalone capability, but the real value comes from using them together. A simple mental model:
– **Work IQ** = how your people work
– **Fabric IQ** = how your business operates
– **Foundry IQ** = how knowledge is connected and reused
– **Web IQ** = what is happening in the outside world
### Work IQ workplace intelligence
Work IQ delivers workplace intelligence built from the signals flowing through Microsoft 365 every day. It continuously builds an up-to-date semantic understanding across documents, meetings, chats, and workflows so agents can navigate the constantly changing context of how your organization actually operates.
In practice, this is the layer that knows who works with whom, what was decided in last week’s meeting, and which document is the latest version. It draws from the tools your people already live in: Teams, Outlook, SharePoint, OneDrive, Word, Excel, PowerPoint, OneNote, and Dynamics.
### Fabric IQ the semantic foundation
Fabric IQ grounds AI in how the business operates. It provides a shared semantic foundation by reusing the definitions, metrics, and relationships already modeled in your data and analytics systems, so agents and people reason from the *same* business meaning rather than conflicting interpretations.
Built on OneLake as its single, governed source of truth, Fabric IQ models business data through ontologies, semantic models, graphs, and data agents. One detail worth calling out for anyone in the data space: Power BI semantic models account for the overwhelming majority of semantic models in production today, and you can bootstrap a Fabric IQ ontology directly from an existing Power BI model in a few clicks. That is a significant accelerator if your business logic already lives in Power BI.
### Foundry IQ governed, reusable knowledge
Foundry IQ is a managed knowledge layer for enterprise data. It transforms fragmented information into governed, reusable knowledge bases that connect agents to the right data, systems, and context.
The key concept here is the **knowledge base**: a reusable, topic-centric collection (think “employee policies” or “product documentation”) that you define once in the Foundry portal. Any number of agents can then connect to that same knowledge base through a single API. No more rebuilding chunking logic, embeddings, routing, and permissions for every project.
Foundry IQ federates across both indexed and remote sources, including Microsoft 365, SharePoint, OneLake, Azure Blob Storage, the web, and MCP servers. When an agent queries it, Foundry IQ uses **agentic retrieval** to plan the query, search across sources, enforce the user’s permissions, and return grounded answers with citations.
### Web IQ fresh, real-world context
Web IQ is the newest addition, announced at Build 2026. It connects agents to fresh, real-world information from across the web: web pages, news, images, and videos. It is built on roughly two decades of large-scale search and retrieval expertise from Bing, re-architected specifically for LLMs and multi-step agents.
Microsoft is positioning Web IQ around grounding quality and speed, citing sub-165 ms P95 latency and token efficiency designed to deliver high-quality answers at lower cost. This is the layer that lets your agents reason about events outside your four walls without you stitching together a web search pipeline.
## How the layers fit together
The point of Microsoft IQ is that these layers compose. An agent does not have to choose one. A single agent can pull workplace signals from Work IQ, ground its reasoning in business semantics through Fabric IQ, retrieve governed knowledge from Foundry IQ, and enrich it all with current web context from Web IQ.
Foundry IQ is interesting in this respect because it can federate the *other* layers as knowledge sources. A Foundry IQ knowledge base can reach into Work IQ, Fabric IQ, OneLake, SharePoint, and the web, then present all of it to an agent through one consistent endpoint.
## Security and governance are built in, not bolted on
For anyone who lives in platform governance, this is the most important section. Microsoft IQ is designed to work *with* the controls you already have rather than introducing a parallel permission model:
– **Identity and permissions are respected.** Agents only access data the requesting user is authorized to see, using your existing Microsoft Entra access controls.
– **Sensitivity and lineage are preserved.** Microsoft Purview sensitivity labels, policies, and data lineage are maintained as AI reasons across systems.
– **Governance applies consistently at scale.** The same rules apply across Copilot, custom agents, and applications, rather than one-off controls per use case.
– **It is observable.** Policy-aware access and observability help you understand and govern how AI uses enterprise data.
If you want a single control plane to observe, govern, and secure every agent across the organization, Microsoft points to **Agent 365** as the companion piece.
## What this means if you build on the Power Platform
Microsoft IQ is not a Power Platform feature, but it sits directly underneath a lot of what we build:
– **Copilot Studio agents** are a primary consumer of this intelligence layer. A point-and-click agent can connect to a Foundry IQ knowledge base instead of you wiring up custom RAG, and it can pull workplace context from Work IQ surfaced through Teams and Microsoft 365 Copilot.
– **Your Power BI investment becomes reusable.** Because Fabric IQ can bootstrap ontologies from existing Power BI semantic models, the business logic and definitions your organization already trusts become the grounding for agents. That curated layer of measures, hierarchies, and dimensions is no longer locked inside reports.
– **Governance scales with you.** The Entra and Purview-based controls you already enforce extend to agents by default, which is exactly the story you want when a business unit asks to push an agent into production.
For makers and platform engineers, the practical takeaway is to stop hand-building context pipelines for every agent. The shared layer exists; the job shifts toward modeling your business well in Fabric IQ, curating reusable knowledge bases in Foundry IQ, and governing the whole estate consistently.
## A quick timeline
– **Ignite 2025:** Work IQ, Fabric IQ, and Foundry IQ introduced as Microsoft’s unified intelligence layer.
– **Build 2026:** Web IQ announced, adding real-time global context from the web and completing the four-layer picture.
## Where to learn more
– Microsoft IQ overview: microsoft.com/ai/microsoft-iq
– Fabric IQ docs: Microsoft Learn (Fabric IQ overview)
– Foundry IQ docs: Microsoft Learn (Foundry agents concepts)
– The Microsoft IQ Series: a hands-on developer program with video episodes, Jupyter notebooks, and Azure deployment templates
## The bottom line
Microsoft IQ is best understood as the answer to one question: how do you give every agent in your organization a shared, governed, real-time understanding of your business so it is useful on day one? Work IQ covers how people work, Fabric IQ models how the business operates, Foundry IQ connects and reuses knowledge, and Web IQ keeps it grounded in the outside world. The names are confusing at first, but the architecture underneath is coherent, and it is built to plug into the Microsoft investments you already have.
