Microsoft just announced something that quietly reframes how enterprise AI agents will get their intelligence. It’s called Work IQ, and the one-line version is this: it’s a workplace intelligence layer that lets agents access and reason over your organization’s data, context, and tools across Microsoft 365 and external systems with permission-aware governance baked in.
If you build agents on the Power Platform or in Copilot Studio, this is worth understanding now, because it changes what “grounding an agent in company data” actually requires.
The shift: intelligence built for agents, not individuals
Until now, most enterprise AI has been designed around a human sitting in front of it — Copilot answering your questions in Word or Teams. Work IQ flips that assumption. It’s built for an agent-first world where AI agents, not developers, dynamically reason, pull context, and act across systems on their own.
The practical promise is that an agent built on Work IQ can operate with the same depth, grounding, and reliability as Microsoft 365 Copilot itself — without you hand-wiring all the retrieval and orchestration plumbing underneath it.
The four things Work IQ unifies
Microsoft frames Work IQ around four capabilities, and they map cleanly to the problems anyone who has built a real agent has hit:
- Chat — conversational intelligence with support for agent-to-agent collaboration (A2A) and REST for web apps, so agents can hand work off to each other, keep continuity, and return fully processed, Copilot-quality answers.
- Context — the part that usually eats your time. Work IQ assembles and grounds context internally across organizational data and hands the agent ready-to-use inputs, so you’re not stitching together raw signals or maintaining retrieval pipelines yourself.
- Tools — a small, composable set of generic tools and skills that let an agent reason, retrieve, and act across Microsoft 365 through a simplified surface, with governance centralized.
- Workspaces — persistent working storage (built on SharePoint Embedded) that lives inside your tenant boundary, giving agents a place to keep intermediate data and outputs for long-running workflows and clean handoffs.
The clever part: 10 tools instead of hundreds
Here’s the piece I find most interesting as an engineer. Instead of exposing hundreds of discrete API operations, the Work IQ MCP server collapses everything into just 10 generic tools that reach Microsoft 365 data — mail, calendar, files, people, chat, and sites — and act on it.
The design treats tools as simple verbs (fetch, create, update) while resource paths describe what the agent is operating on. That separation keeps the surface tiny but flexible, so new data sources don’t require new tools.
Tying it together is a capability called getSchema, which lets an agent discover how data is structured at runtime rather than relying on predefined models. In effect, every data source becomes self-describing — the agent learns what exists and how to work with it on the fly, and the API surface doesn’t have to change as scenarios evolve. For anyone who has maintained brittle integrations against shifting schemas, that’s a meaningful idea.
Security and governance, centralized
External-facing intelligence is only as good as its guardrails, and this is where Work IQ takes a notably different approach. Rather than juggling hundreds of static OAuth scopes, it uses a small set of broad permissions to set high-level boundaries, then enforces detailed, context-aware rules on every single request through a Rego-based policy engine(the same policy language behind Open Policy Agent).
Two things stand out:
- Every action is user-scoped. A request runs in the context of a specific user and can only touch what that user is allowed to see or do — the agent doesn’t get to bypass a person’s permissions.
- Policy is evaluated per request, factoring in resource paths, request methods, identity, and even data content. That lets an organization enforce rules like blocking sensitive-data access, preventing exfiltration, or limiting external communication — without changing the API.
On top of that, observability is built in: every tool invocation is logged and evaluated, enabling auditing, usage analytics, rate limiting, and real-time compliance across agents and data sources. For governance-minded teams, “auditable by default” is a strong starting position.
Who’s already building on it
The announcement points to real production usage, not just demos. A few examples, paraphrased:
- SLB is using Work IQ inside its TELA agentic assistant to blend subsurface engineering data with Microsoft 365 context — meetings, documents, communications — so analysis workflows can draw on broader organizational knowledge, with results flowing back into Teams, email, and SharePoint.
- HP is integrating it into AI-enabled multifunction printers, so a user can scan a document and have it summarized, translated, redacted, and pulled into agent workflows right from the device panel.
- Miro is bringing Microsoft 365 context — emails, docs, chats, meeting activity — onto its collaborative canvas, so its AI features operate on the same picture the rest of the org is working from.
Partners named alongside these include Cisco, darwinbox, and kor.ai, among others.
Availability and cost — the practical details
A few specifics worth noting if you’re planning around this:
- GA is June 16, 2026, bringing the Work IQ API endpoints: A2A, a redesigned remote MCP server, and a REST API. It’s already in public preview ahead of that.
- Access is independent of Microsoft 365 Copilot licensing and offered on a consumption model. Copilot-licensed users get Work IQ across Copilot experiences and agents, with consumptive billing for custom and third-party agents; users without a Copilot license are billed consumptively as well.
- Cost is variable based on usage and will be manageable in the Microsoft 365 admin center starting mid-June, alongside new enterprise governance and cost-management controls.
As always with brand-new, consumption-priced services, confirm the current licensing and pricing details before you commit a production budget — this is rolling out as of this writing.
What this means if you build on the Power Platform
For Copilot Studio and Power Platform builders, the signal here is that Microsoft is standardizing the hard part — grounded, governed, permission-aware access to organizational data — into a reusable layer with a tiny tool surface and runtime schema discovery. If you’ve been assembling your own connectors, retrieval logic, and permission handling to give an agent real organizational context, Work IQ is aiming squarely at that work.
It’s early, it’s consumption-priced, and it’s worth getting hands-on with in preview before GA so you understand the tool model and governance approach before it shows up in production scenarios you’re responsible for.
Based on Microsoft’s announcement, “Work IQ: Production-ready intelligence for every agent” on the Microsoft 365 Developer Blog (June 2, 2026). For technical details and samples, see the Work IQ overview on Microsoft Learn and the Work IQ GitHub repo.