What it does
Work IQ is the intelligence layer that sits under Microsoft 365 Copilot and its agents. It grounds them in your organisation's work data and keeps a memory of how people and teams actually work, then passes that context to Copilot and to agents, including agent-to-agent handoffs.
Key facts
- Announced at Ignite 2025 (November 2025) as the reasoning and grounding layer for Copilot and agents.
- Two parts: your work data (emails, files, meetings, chats, calendar) that captures how work gets done, and memory of style, preferences, habits, workflows and working relationships.
- Uses these signals to match a request to the right agent and to give role-aware context rather than a cold prompt.
- Work IQ for custom agents lets you build agents tuned to your workflows, grounded in the same layer, through Copilot Studio or via API for pro-code work.
- Grounding respects existing permissions, sensitivity labels, compliance controls, audit, logging, monitoring and policy enforcement.
- Sits alongside Fabric IQ (business entities and data in Microsoft Fabric) and Foundry IQ (managed knowledge grounding) as the three "IQ" layers.
- Shares context across agents, so a delegated agent inherits relevant working context instead of starting blind.
- Part of the wider Agent 365 and "Frontier Firm" positioning; some capabilities are still rolling out through 2026.
When to use / skip
You don't switch Work IQ on and off; it's the substrate Copilot runs on. The real decision is whether to build custom agents on top of it. Do that when a workflow is repetitive, org-specific and benefits from knowing the person and their recent work. Skip the custom-agent route for one-off tasks where stock Copilot chat already has the context in front of it.
Configuration decisions
- Decide which content Work IQ can reach through Microsoft 365 permissions and sensitivity labels before building agents on it, not after.
- Choose Copilot Studio (low-code) or the API (pro-code) for custom agents depending on who owns the build.
- Confirm your audit, logging and policy tooling covers agent activity, since agents inherit these controls but someone has to watch them.
- Work out where Work IQ ends and Fabric IQ or Foundry IQ begins for grounding on structured business data or curated knowledge.
Gotchas
- Work IQ is only as good as the underlying data hygiene and permissions; oversharing at the SharePoint layer becomes oversharing at the agent layer.
- Memory of "how you work" raises reasonable questions from users and works councils about what is being retained; get ahead of it.
- The three IQ layers are easy to conflate in a pitch; be precise about which one grounds which kind of data.
- Much of the custom-agent and agent-to-agent context sharing landed in late 2025 and is still maturing, so check current availability before committing a design.
Consultant notes
- Position Work IQ as plumbing, not a feature to sell. The client-visible value shows up in agent quality and Copilot relevance.
- Data governance readiness is the gating factor for any Work IQ story. Lead with a permissions and sensitivity-label review, not with agent demos.
- For clients already nervous about Copilot grounding, Work IQ is a chance to show controls (labels, audit, policy) rather than to gloss over them.
- Tie it to the Agent 365 conversation early; clients who want fleets of agents need to understand what shared context buys them and what it exposes.
Check current availability of Work IQ for custom agents and agent-to-agent context before you design around it; this moved fast through 2026.