Easiest way to hold these in your head is as three layers answering three different questions about your organization.
Work IQ, how work happens. Collaboration signals, meetings, messages, documents, project dynamics, and the relationships between people. It sits inside the Microsoft 365 trust boundary and is the intelligence engine behind Microsoft 365 Copilot. Use it when an agent needs human and organizational context, who owns something, what was discussed, what changed recently.
Fabric IQ, how the business operates. The semantic layer over your structured and semi-structured data. It defines shared metrics and business ontologies so agents interpret Revenue or Customer consistently rather than each agent guessing. Note the boundary: it covers analytical data only. It does not read human communications and does not crawl PDFs or policy documents.
Foundry IQ, what the organization knows. The governed knowledge retrieval layer on Azure, spanning Fabric, Microsoft 365, and other enterprise content. This is the one that replaces hand-built RAG pipelines. Without it, every AI use case becomes its own retrieval engineering exercise with duplicated logic and inconsistent results, which anyone who has built a few RAG pipelines will recognize.
There is also Web IQ for external web context, which completes the set.
The use case that made it click for me
Supply chain delays, and it is the example Microsoft themselves use. Fabric IQ detects the anomaly in delivery metrics, spotting suppliers trending late beyond historical norms and on-time percentages dipping in specific regions. Foundry IQ retrieves the supplier contracts, SLAs, and penalty clauses, so the agent understands the contractual obligations around those delays. Work IQ supplies the human context, who has been raising this in meetings and what has already been escalated.
Any one layer alone gives you a partial answer. A dashboard tells you deliveries are late. The three together tell you which supplier, what the contract entitles you to, and who is already dealing with it.
A second, simpler framing: an agent asked how the pipeline is trending against last year's Q4 and whether any delays are flagged in customer conversations. Structured pipeline data is Fabric IQ, conversation signals are Work IQ, and Foundry IQ synthesizes across both with consistent semantics.
Practical points if you are evaluating this
Start with one use case rather than adopting all the layers at once. Map which layers your agent actually needs before building, since many scenarios only need one.
Governance first. These layers make considerably more organizational data queryable by agents than before, so pair any rollout with a data access review. Agent 365 is the governance and observability layer that sits alongside this.
Work IQ billing activation has to be switched on by your M365 tenant admin, so budget for it rather than assuming it is included.
Worth understanding how these relate to the build surfaces too. Microsoft Foundry is the pro-code environment, Copilot Studio is low-code business agents, and Agent 365 is the runtime governance layer. The IQ layers are what any of those ground against.
For video, John Savill has a good overview covering model knowledge, enterprise data types, use cases, and agent integration, organized into short chapters. Worth an hour if you want the demos rather than the concepts.
