I've spent the better part of the last eighteen months watching enterprise AI agent projects hit the same wall. Agents that reason well in the abstract stumble badly when you ask them about your business. Who owns this account? What was decided in Thursday's architecture review? Has anyone followed up on the contract renewal? These are not hard questions for a human employee with six months of context. For a model working from a system prompt and a handful of retrieved chunks, they're nearly intractable.
Today, Microsoft ships something that takes a serious swing at that problem. The Work IQ API reaches general availability on June 16, 2026, and if you build enterprise software or AI systems on top of Microsoft 365, it deserves more than a quick skim of the announcement post.
What Work IQ Actually Is
The description that keeps circulating — "it gives agents access to your Microsoft 365 data" — undersells what's here and misses the important nuance. You can already pull email through the Graph API, query SharePoint documents, read calendar events. What Work IQ does differently is sit one level above that raw data layer and maintain a semantic understanding of how your organization actually operates.
According to Microsoft's announcement, the average Fortune 500 company carries over 600 terabytes of data across their Microsoft 365 tenant — email threads, meeting recordings, chat histories, document drafts, collaboration patterns, signals from line-of-business systems. Work IQ continuously processes that corpus into structured organizational context: who knows what, what decisions have been made and when, what the active projects and owners are, how teams collaborate. When you call the API, you're not running a keyword search against raw files. You're querying a continuously-updated intelligence layer that surfaces the business context relevant to your query. That distinction matters when you're building agents that need to make sound decisions rather than simply retrieve documents.
The Four API Domains
The GA release ships four distinct domains:
- Chat — Programmatic access to Microsoft 365 Copilot's full reasoning capabilities. Returns responses with citations, supports agent-to-agent delegation, and maintains conversation continuity. Think of it as an authenticated, auditable way to invoke Copilot reasoning from inside your own agent workflows.
- Context — Delivers pre-assembled, agent-ready context packages in token-efficient formats, rather than requiring your agent to retrieve and stitch together raw documents. Microsoft claims 80% fewer tokens consumed versus traditional Graph API approaches for equivalent context depth. For teams with token costs on the income statement, that is not a rounding error.
- Tools — Ten generic action verbs — fetch, create, update, and similar primitives — mapped to resource paths across Microsoft 365. Send email, schedule meetings, upload to SharePoint, post to Teams. The surface area is deliberately small while covering a wide action space. A standout feature is
getSchema, a tool that lets an agent dynamically discover how organizational data is structured at runtime, without requiring you to encode schema assumptions at build time. For agents operating across multiple tenants or organizational units with different configurations, this matters. - Workspaces — Persistent state storage for agents, backed by SharePoint Embedded, that never leaves your tenant boundary. Agents can park intermediate results, working memory, and progress checkpoints across multi-step workflows without those artifacts touching external systems.
Three Integration Paths: MCP, A2A, and REST
The GA release ships three distinct ways to connect, and the choice between them reflects where the industry is heading.
The REST API is the baseline — authenticated HTTP calls, standard JSON, nothing novel. It's the path for teams already embedded in the Microsoft ecosystem who need straightforward integration into existing services.
More telling is the redesigned remote MCP server. The Model Context Protocol has been gaining real traction across AI tooling as a standardized way for models and agents to discover and invoke external capabilities. Microsoft shipping a production-grade remote MCP endpoint for Work IQ is a significant endorsement of MCP as an interoperability standard. It means any agent runtime that speaks MCP — Claude Code, Cursor, custom orchestration frameworks — can wire into Microsoft 365 organizational context without building a bespoke connector. For teams already running MCP-based tool discovery, this is immediate.
The third path is Agent-to-Agent (A2A), which supports delegation patterns where one agent passes context and task state to another. Multi-agent architectures have moved from experimental to standard over the past year, and most teams I know are either building delegation infrastructure themselves or bolting it on after the fact. Work IQ ships A2A delegation natively at GA — that's one less piece of plumbing you're writing yourself.
The Part That Matters Most: Governance
I want to be direct: the governance model is the most important part of this announcement for anyone running production systems.
Work IQ executes all requests within a specific user context. Every API call is scoped to what that user is authorized to access — the API will not surface data the user couldn't otherwise see. That's a critical property for enterprise deployments. The alternative is either granting agents overly broad permissions or building your own authorization middleware to enforce per-user scoping — both expensive choices.
Policy enforcement uses a Rego-based rules engine — the same policy language used by Open Policy Agent. If your team already manages access control with OPA, the mental model transfers directly. Context-aware rules can adjust what the API surfaces per request, not just at configuration time but dynamically at query time.
Audit logging and usage analytics ship in the box, surfaced through the Microsoft 365 admin center. For teams under active SOC 2 or ISO 27001 scrutiny, the ability to answer "what data did the agent access, when, and on whose behalf" is a hard requirement. Historically you'd build that instrumentation yourself or accept the gap. Work IQ closes it.
Pricing: What the Copilot Credits Model Actually Means
Work IQ usage is billed through Copilot Credits — the same consumption currency used by Copilot Studio and other Microsoft AI services — and it is explicitly independent of Microsoft 365 Copilot seat licensing. There is no separate Work IQ API subscription or per-user SKU. You pay on actual consumption.
There are two billing components: a fixed per-call cost for Tools invocations, and a variable cost for Chat and Context that scales with the complexity of context assembled. The variable component ties directly back to the token efficiency claim — if Context API genuinely delivers 80% token reduction versus raw Graph API calls, cost per interaction should be meaningfully lower than a first-pass estimate would suggest.
The licensing decoupling is the more interesting architectural choice. ISVs and startups can build products on Work IQ without requiring their enterprise customers to already hold Copilot seat licenses. That opens a much larger addressable market for third-party agents and removes a sales blocker that would otherwise slow adoption across the partner ecosystem.
Who Should Pay Attention Today
If you build agent systems for enterprises running Microsoft 365 — which globally is hundreds of millions of licensed seats — today's GA release gives you a production path to organizational context that would otherwise require months of custom Graph API integration work. Real deployments are already running: SLB is using Work IQ to connect subsurface domain data with enterprise workflow context inside agentic pipelines. HP is running document scanning and translation workflows against it. Miro is surfacing real-time Microsoft 365 context on collaborative canvases to support AI features.
If you're on the platform or infrastructure side, focus your evaluation on the governance controls. The Rego policy engine, user-scoped execution, tenant-boundary storage via SharePoint Embedded, and built-in audit logging are the features that will determine whether your security team clears this for production. They're asking exactly these questions now.
If you're a smaller team or building a proof of concept, the MCP path is your fastest on-ramp. Connect to the remote MCP server, configure your agent runtime, and you have organizational context in your stack in far less time than a full Graph API integration would require.
My Take
I've watched too many enterprise AI rollouts stall because agents couldn't get the context they needed without someone engineering a custom integration layer for every data source. The manual retrieval approach doesn't scale. Neither does the "throw everything in the system prompt" pattern when you're looking at 600 TB per tenant.
Work IQ is Microsoft's answer at platform scale. It won't solve everything — you'll still need to design your agent logic carefully, think hard about which API domain suits each task, and manage credit consumption as usage grows. But the architecture reflects real thinking about how teams actually ship and operate agent systems: MCP for interoperability, A2A for multi-agent delegation, Rego for policy, user-scoped execution for security, consumption-based pricing for market reach.
Those are choices I'd have made. The GA release is live today. If you've been waiting for Microsoft 365 organizational context to graduate from preview territory to something you can ship against in production, today is the day that happened.