Microsoft Work IQ Preview: Unified Enterprise Context for Copilot Agents Arrives November 2026

Microsoft is continuing to build out its answer to that exact problem. A newly published entry on the Microsoft 365 AI at Work Roadmap—tracked under Roadmap ID 570870 and titled "Work IQ: Microsoft IQ"—outlines an upcoming release aimed at giving AI agents a shared brain for corporate knowledge. Currently listed as "In development," the feature is slated for a public preview in November 2026, followed by a targeted General Availability (GA) rollout in January 2027 for Worldwide (Standard Multi-Tenant) cloud instances on the web.

Before anyone marks their calendar in permanent ink, the usual Microsoft roadmap caveat applies: roadmap timelines are estimates subject to change, and Microsoft routinely shifts dates as features move through testing. Still, the entry signals the next phase of Microsoft’s push to turn Work IQ into the central nervous system for Microsoft 365 Copilot and custom enterprise agents.

The Roadmap Addition vs. What Already Exists​

To understand why Roadmap ID 570870 matters, it helps to separate this specific upcoming milestone from the broader Work IQ ecosystem Microsoft has been rolling out over the past several months.

Work IQ itself is not brand new. Microsoft introduced the underlying Work IQ intelligence layer earlier, announced the new Work IQ APIs at Microsoft Build on June 2, 2026, and brought Work IQ APIs into general availability on June 16, 2026. In late September and early October 2026, Microsoft also detailed how Work IQ grounds Microsoft Copilot and agents in business data across Dynamics 365 and the Power Platform, alongside an official GitHub Copilot plugin collection and CLI for Work IQ.

So what is Roadmap ID 570870? Tagged under both Work IQ and Microsoft Copilot (Microsoft 365) for the web platform, the entry describes a capability to "Provide a unified intelligence layer that helps agents discover, reason over, and use organizational knowledge."

While the terse roadmap blurb does not spell out every granular UI change or backend hook arriving in November, its placement under the "Microsoft IQ" banner points to a formalization of how agents tap into enterprise-wide knowledge graphs directly through the web experience. For IT administrators and developers, the value here isn't just watching a roadmap number tick toward preview; it is understanding how the Work IQ architecture actually operates under the hood—and what governance levers you need to pull before those agents start poking around your tenant.

Summary: Roadmap ID 570870 targets a November 2026 preview and January 2027 general availability for "Work IQ: Microsoft IQ" on the web, expanding Microsoft's existing workplace intelligence framework.

Under the Hood: How Work IQ Actually Works​

If you strip away the marketing gloss, Work IQ is essentially an abstraction and governance layer designed to stop developers from having to build custom retrieval-augmented generation (RAG) pipelines and tangled OAuth permission trees for every single bot they deploy.

According to Microsoft Learn documentation, Work IQ is structured around three foundational pillars—Data, Memory, and Inference—and exposes its capabilities to agents through four functional domains: Chat, Context, Tools, and Workspaces.

1. The Three Layers: Data, Memory, and Inference​

• Data: Connects signals across unstructured Microsoft 365 productivity assets (emails, Teams chats, SharePoint files, meetings) and structured business workflows in Dynamics 365 and Power Platform.
• Memory: Maintains a persistent understanding of how specific users and teams operate, keeping agents aligned across different sessions and applications rather than resetting to zero after every prompt.
• Inference: Pairs foundation models (including OpenAI and Anthropic models supported within Microsoft 365 Copilot) with tools so agents can reason over data and execute multi-step actions under the supervision of the Agent 365 control plane.

2. The Four Functional Domains​

When developers wire an agent up to Work IQ—whether via Agent-to-Agent (A2A) protocols, a remote Model Context Protocol (MCP) server, or standard REST APIs —they interact with four core building blocks:

• Chat: Handles conversational continuity and A2A delegation, allowing one specialized agent to hand a task off to another without losing the thread of the user's original request.
• Context: Eliminates custom orchestration overhead by pre-assembling and grounding organizational data into agent-ready inputs.
• Tools (The 10-Verb Simplification): Historically, connecting an AI to Microsoft Graph meant wrestling with hundreds of specific API endpoints and static OAuth scopes. Microsoft Learn notes that the Work IQ MCP server collapses hundreds of operations into just 10 generic tools functioning as simple verbs (such as fetch, create, and update) paired with resource paths across mail, calendar, files, people, chat, and sites. Even better, agents can dynamically discover how data is structured at runtime, treating data sources as self-describing interfaces rather than breaking whenever a schema changes.
• Workspaces: Long-running agent workflows need a scratchpad. Work IQ uses SharePoint Embedded working storage to give agents a persistent, secure space inside the tenant boundary to store intermediate data and hand off results between steps.

Summary: Rather than forcing each custom agent to build its own data retrieval pipeline, Work IQ unifies Microsoft 365 and Dynamics 365 signals behind a 10-tool MCP surface, runtime schema discovery, and SharePoint Embedded workspaces.

Security, Rego Policy Enforcement, and the "Read-Only" Catch​

Giving autonomous or semi-autonomous agents a unified way to "discover, reason over, and use organizational knowledge" naturally raises the hair on the back of every CISO’s neck. What stops an overly helpful agent from discovering the CEO’s M&A folder or accidentally overwriting a production SharePoint list?

Microsoft’s architectural answer relies on user-scoped execution and centralized policy enforcement:

1. User-Scoped Context: Work IQ does not operate as an all-seeing god account. Every request executes strictly within the identity context of the specific user invoking the agent, meaning the agent can only see or act upon data that the human user is already permitted to access.
2. Rego-Based Policy Engine: Instead of managing hundreds of brittle OAuth scopes, the Work IQ MCP establishes high-level access boundaries and runs every single request through a Rego-based policy engine. This engine evaluates the resource path, request method, user identity, and actual data content in real time before allowing an action to proceed.
3. Read-Only by Default: According to Microsoft’s Copilot Studio documentation, Microsoft 365 Work IQ is strictly read-only by default. Agents cannot create, modify, or send items unless an administrator explicitly enables write operations inside the Microsoft 365 admin center.
4. Full Observability: Every tool invocation is logged and evaluated for compliance, rate limiting, and auditability.

The Licensing and Billing Reality Check​

IT leaders budgeting for the late 2026 and early 2027 rollouts also need to pay close attention to how Work IQ is metered.

According to Microsoft Learn, while Work IQ is included across first-party Microsoft 365 Copilot experiences for users who hold a standard Microsoft 365 Copilot license, Work IQ API access is independent of Microsoft 365 Copilot licensing and runs on usage-based billing.

• If a licensed Copilot user interacts with custom or third-party agents tapping into Work IQ, usage-based billing applies.
• If an unlicensed user interacts with Work IQ-powered agents, usage is billed on a consumption basis.
• In Copilot Studio (where the Work IQ preview utilizes the GitHub Copilot harness), consumption is metered via Copilot Credits, and administrators must configure a separate spending policy in the Microsoft 365 admin center to govern costs.

Summary: Work IQ enforces user-scoped permissions via a Rego policy engine and defaults to read-only access in Microsoft 365 until an admin flips the switch on write operations. Meanwhile, custom agent access requires usage-based billing policies.

How to Test Work IQ in Copilot Studio Today​

While we wait to see what specific web capabilities light up when Roadmap ID 570870 hits preview in November 2026, enterprise developers and admins don't have to sit on their hands. You can already test the underlying Work IQ (Preview) integration inside Microsoft Copilot Studio right now to see how the unified MCP tool surface behaves in a non-production environment.

(Note: Microsoft warns that the Copilot Studio integration is a preview feature governed by supplemental terms, is not intended for production workloads, and requires an active spending policy for Copilot Credits. Additionally, admin center controls to allow or disallow MCP servers may not yet be visible in all regions.)

If your tenant is set up for testing, Microsoft Learn outlines the following workflow to wire an agent to Work IQ:

1. Open Copilot Studio: Navigate to your test environment and either create a new agent or select an existing one.
2. Access the Tool Catalog: Select the Tools tab in your agent configuration, then click Add Tool.
3. Select the MCP Surface: On the Add a tool screen, filter by MCP to view available Model Context Protocol servers, and select Work IQ (Preview). (Crucial tip from Microsoft's documentation: Do not select the legacy, workload-specific tools like separate Mail, Teams, or SharePoint connectors if you want the unified Work IQ runtime—those use the older harness and billing model.)
4. Authenticate the Connection: Expand the Connection dropdown, select Create new connection, click Create, and sign in with your test tenant credentials.
5. Configure and Test: Click Add and Configure. Once attached, open the Test pane and feed the agent a prompt that requires live organizational context (for example, asking it to check the status of a project or draft a test email if write operations are enabled in your sandbox). When prompted by the interface, select Allow to grant Work IQ permission to connect to your services.

What IT Admins Should Watch Heading Into November​

Roadmap ID 570870 ("Work IQ: Microsoft IQ") is another clear indicator that Microsoft views the raw LLM war as yesterday's news; the real enterprise battleground is context orchestration. By allowing organizations to swap between OpenAI and Anthropic models while keeping the underlying organizational memory, permission boundaries, and 10-verb MCP toolset constant, Microsoft is trying to make Work IQ the indispensable middleware of corporate AI.

As the November 2026 preview window approaches, Windows and Microsoft 365 administrators should keep three practical questions on their radar:

1. What distinguishes "Microsoft IQ" on the web? Will Roadmap ID 570870 introduce a dedicated web portal or visual knowledge graph for exploring organizational context, or is it primarily a backend upgrade to how web-based Copilot agents discover tenant knowledge?
2. Are your SharePoint and Dataverse permissions clean? Because Work IQ dynamically discovers data structures at runtime within a user's permission scope, any overshared SharePoint site or loosely permissioned Dynamics 365 table is immediately discoverable by an agent acting on that user's behalf.
3. Are your spending guardrails in place? With custom agents drawing on usage-based billing and Copilot Credits, setting up cost management policies in the Microsoft 365 admin center before enabling tenant-wide agent tools will prevent nasty surprises on your monthly Azure and M365 invoice.

Are you already experimenting with the Work IQ MCP server in Copilot Studio or the GitHub Copilot CLI, or are you keeping AI agents locked in read-only mode until 2027? Drop your thoughts and sandbox experiences in the WindowsForum.com community threads below!