Microsoft gives AI agents the keys to Microsoft 365 context

Building an AI agent that touches corporate data has always been a painful plumbing job. Developers usually had to pull raw emails, chats, and calendar invites through standard APIs, stitch everything into an unwieldy prompt, and pray the model wouldn't choke on context limits.

Microsoft wants to change that. On June 2, the company introduced its new Work IQ APIs, pitching them as a dedicated semantic intelligence layer for Microsoft 365. The APIs are already in public preview on GitHub and will be generally available on June 16, 2026.

Collapsing enterprise bloat into 10 generic tools

Traditional enterprise APIs were built for humans clicking buttons, not for autonomous agents performing high-frequency actions. Work IQ continuously crunches emails, meetings, chats, documents, and corporate org charts, serving back pre-digested business context rather than raw data dumps.

The performance gains are substantial. Microsoft claims the Work IQ APIs deliver 2x faster processing speeds and slash token usage by 80% compared to traditional approaches. By moving heavy data packaging and ID trimming into the Work IQ runtime, agents get straight to reasoning instead of parsing JSON.

Instead of forcing developers to register hundreds of specialized data endpoints, Work IQ collapses tool calling into just 10 generic tools via the Model Context Protocol (MCP) using progressive disclosure. The API is divided into four domains: Chat for full Copilot synthesis, Context for structured source data, Tools for execution, and Workspaces, which acts as a secure staging area where long-running agents like Copilot Cowork can stash intermediate states and memory.

Pricing also shifts away from per-seat enterprise licensing. The APIs will run on consumption-based Copilot Credits, charging a flat rate for Tools and variable rates for Chat and Context, backed by spending caps in the Microsoft 365 admin center.

Why it matters

For enterprise developers and AI startups, this removes the hardest layer of the enterprise stack. You no longer have to build your own fragile retrieval pipelines, parse permission trees, or maintain hundreds of API endpoints. It makes building real-world corporate agents dramatically faster, though teams should test their credit burn rates thoroughly before going full auto.

Strategically, this is Microsoft securing its moat against model builders. Foundation models are turning into interchangeable commodities, but real business context remains locked inside enterprise collaboration suites. By turning Microsoft 365 into an agent runtime layer, Microsoft ensures that regardless of which frontier model an enterprise chooses, it still pays the Microsoft toll to get actual work done.

We think the move to adopt MCP and offer persistent workspace memory directly inside the tenant boundary is particularly sharp. It solves the biggest headaches for agent developers: token bloat, state management, and enterprise compliance. The downside? Relying on Work IQ ties your company's agent architecture even deeper into the Microsoft walled garden, making any future cloud migration virtually unthinkable.