Microsoft Work IQ APIs: The Agent Context Layer (2026)

Microsoft turned on the Work IQ APIs on June 16, 2026 — a new way for AI agents to read and act on Microsoft 365 data without stitching together raw data from dozens of separate endpoints. Instead of hundreds of data-specific operations, agents get four domains and ten generic tools exposed over the Model Context Protocol, billed in Copilot Credits. Here's what's actually verified, straight from Microsoft's own announcement.

TL;DR

• The Work IQ APIs became generally available on June 16, 2026, after being announced at Microsoft Build 2026 on June 2.¹
• Work IQ is the intelligence layer behind Microsoft 365 Copilot: it continuously processes email, calendar, meetings, chats, files, people, and line-of-business systems to build a real-time semantic model of how an organization operates.¹
• The API is organized into four domains — Chat, Context, Tools, and Workspaces — and collapses tool calling into just 10 generic tools surfaced through the Model Context Protocol (MCP).¹
• Microsoft's internal testing claims the APIs run 2x faster than traditional APIs and use 80% fewer tokens in coding harnesses; it pegs the average Work IQ data footprint in a Fortune 500 org at over 600 terabytes.¹
• Pricing is consumption-based in Copilot Credits: a fixed component for Tools, variable for Chat and Context, with a new cost-management dashboard in the Microsoft 365 admin center.¹
• You can build against it today through the public preview on GitHub (github.com/microsoft/work-iq).¹

What You'll Learn

• What Work IQ is and where it sits in Microsoft's "IQ" context-layer family
• The four API domains and what each one returns
• Why Microsoft collapsed the surface to 10 generic tools over MCP
• How Work IQ differs from calling Microsoft Graph directly
• The performance and efficiency numbers Microsoft published — and how to read them
• How Copilot Credits pricing works and where to watch your spend

What Work IQ Is

Work IQ is what Microsoft calls "the intelligence layer behind how work gets done."¹ Rather than being a model, it is a context engine: it continuously processes content from email, calendar, meetings, chats, files, people, collaboration patterns, and line-of-business systems, then builds a semantic understanding of the business — "a real-time model of how your organization operates," in Microsoft's framing.¹

That layer already powers Microsoft 365 Copilot. What changed on June 16 is that Microsoft opened it up: the Work IQ APIs let developers bring the same understanding into their own agents, "so they can work with business context, not just raw data, and take actions across your organization."¹ The announcement came from Charles Lamanna, Microsoft's EVP for Copilot, Agents, and Platform, at Build 2026.¹

Work IQ is one piece of a broader umbrella Microsoft introduced at Build called Microsoft IQ — a context layer, generally available across GitHub Copilot, Microsoft Foundry, and Copilot Studio, that grounds agents in both world knowledge and enterprise knowledge.² The family has four modules: Work IQ (workplace intelligence — how work happens across Microsoft 365), Fabric IQ (a shared semantic foundation over structured business data), Foundry IQ (which ties the others together and handles retrieval planning across enterprise knowledge and the live web), and the newly announced Web IQ (real-time grounding from the open web).² Work IQ is the module focused specifically on the day-to-day Microsoft 365 work graph.

The Four API Domains

Microsoft says the Work IQ APIs "consist of four domains that map how agents work."¹ Each one answers a different need an agent has:

The split is deliberate. Chat is for when you want Copilot's finished answer. Context is for when your own model is doing the reasoning and just needs the right grounding material. Tools is for taking action. And Workspaces exists because long-running agents — like Microsoft's own Copilot Cowork and the newly announced Microsoft Scout — need somewhere durable to keep working state between steps.¹

Ten Tools, Not Hundreds

The headline design decision is on the Tools side. Traditional enterprise APIs expose hundreds of data-specific operations, and an agent has to be taught each one. Microsoft instead "collapses operations into just 10 generic tools with progressive disclosure through Model Context Protocol (MCP), so developers do not need to teach agents hundreds of data-specific tools."¹

This matters for anyone building agents. A smaller, stable tool surface means fewer tokens spent describing tools, fewer brittle integrations to maintain, and less prompt real estate burned on schemas the model rarely uses. Progressive disclosure over MCP lets the agent pull in detail only when it needs it. Microsoft frames the payoff as agents moving "from reasoning to action much faster," with higher throughput.¹

Work IQ vs. Calling Microsoft Graph

If you've built on Microsoft 365 before, the obvious question is how this differs from Microsoft Graph. Graph is the long-standing, permission-aware API surface that exposes emails, files, calendars, and relationships as discrete resources you enumerate and stitch together yourself.

Work IQ is positioned a layer above that. Independent technical write-ups describe it as sitting on top of Graph's data access and adding the intelligence and orchestration: a semantic index of organizational content, memory of user and agent activity, and real-time, user-scoped permission evaluation, exposed as a single surface an agent can call.³⁴ Microsoft's own framing is that, instead of an orchestration layer reading raw data and stitching it together, "specialized large language models and agents within Work IQ package relevant context and data in a structure that is easier for agents to consume."¹

The practical guidance that follows: Graph is still the right tool when you need to enumerate specific resources programmatically — list every item in a SharePoint list, read structured calendar entries, hit admin endpoints. Work IQ is the better fit when an agent needs to understand organizational context and act on it, rather than page through raw records.³ They are complementary, not a one-for-one replacement.

The Numbers Microsoft Published

Microsoft attached three efficiency claims to the launch, all from its own internal testing — worth keeping that attribution in mind:¹

The speed and efficiency gains come from doing more processing inside the Work IQ runtime: an agent-optimized retrieval system that cuts round trips, plus trimming of file record strings, message IDs, and app IDs during tool calls so fewer tokens are spent on plumbing.¹ These are vendor figures from controlled internal tests, not independent benchmarks, so treat them as directional rather than guaranteed for your workload.

The 600 TB figure is the more strategic point: it is Microsoft's argument for why a dedicated layer is needed at all. The company is explicitly designing for a world where "hundreds of millions of agents come online over the next few years," each running continuous, high-frequency, multi-step operations rather than the intermittent, shallow access patterns humans produce.¹

Security and the Tenant Boundary

A recurring concern with agent platforms is where data and actions live. Microsoft's answer is that with Work IQ, "data, context, and insights stay within the Microsoft 365 tenant trust boundary, and actions that an agent takes are auditable and discoverable."¹ The Workspaces domain — where agents store intermediate state — is likewise inside that boundary.¹ Microsoft's pitch to IT is that you get enterprise controls "without adding a separate governance layer." As always with security claims, the right move is to validate the audit trail and permission behavior against your own compliance requirements before granting agents write access.

How Pricing Works

Work IQ uses consumption-based pricing denominated in Copilot Credits, with a fixed component for Tools and a variable component for Chat and Context.¹ API access is independent of Microsoft 365 Copilot licensing: users without a Copilot license are billed consumptively, and custom or third-party agents are billed consumptively even for licensed users.⁴ To manage that spend, Microsoft is launching a cost-management dashboard in the Microsoft 365 admin center where IT can review AI credit usage, choose between prepaid and pay-as-you-go billing, and set spending limits across tenants, groups, and users. Work IQ APIs are the first product managed through that dashboard, with more Copilot Credit products to follow.¹

For exact rates, Microsoft points to its Work IQ licensing page rather than publishing per-credit numbers in the announcement, so confirm current pricing there before you model costs.¹ The headline structure to plan around: actions (Tools) carry a predictable fixed cost, while reasoning-heavy Chat and Context calls scale with usage.

A Note on the Code

Work IQ exposes its capabilities in three formats — Agent-to-Agent (A2A) for agent collaboration, MCP for IDEs and tools, and a REST API for app integrations — and the public preview lives on GitHub at github.com/microsoft/work-iq.¹⁴ Because the exact request and response schemas are still settling around the June 16 GA, this post deliberately avoids reproducing endpoint signatures that could drift — pull the current schemas from the GitHub preview and the official docs rather than copying call examples from third-party blogs. Conceptually, though, the mental model is simple: point your agent's MCP client at the Work IQ tool server, and it discovers the ten generic tools — simple verbs like fetch, create, and update paired with resource paths, plus a getSchema call to learn how data is structured at runtime; call Context to ground your own model, Chat to get Copilot's finished answer, Tools to act, and Workspaces to persist state across steps.

If you're wiring agents into Microsoft's broader stack, Work IQ slots in next to Microsoft Agent 365, the control plane for managing agents, and it follows the same MCP-first direction we covered in GitHub Copilot's roadmap. For a contrast in how another platform vendor is exposing on-device and server models to developers, see our look at Apple's Foundation Models framework.

The Bottom Line

Work IQ is Microsoft's bet that the next wave of software is agents, not apps — and that agents need a context layer, not just a data API. Collapsing Microsoft 365 access into four domains and ten MCP tools is a genuinely developer-friendly move, and the efficiency story (fewer tokens, fewer round trips, state that lives inside the tenant) is aimed squarely at the pain of building production agents today. The asterisks are the usual ones: the 2x-faster and 80%-fewer-tokens figures are Microsoft's own internal numbers, Copilot Credits pricing means you'll want that admin dashboard on from day one, and the API schemas are fresh enough that you should build from the GitHub preview rather than secondhand examples. If you're already invested in Microsoft 365, Work IQ is the most direct path yet to giving your agents real organizational context.

This topic is moving quickly. All figures here are sourced to Microsoft's official announcement and Build 2026 materials as of June 19, 2026; pricing, schemas, and availability can change.