Microsoft just gave its AI agents a memory upgrade. At its Build 2026 conference on June 2 in San Francisco, the company introduced Microsoft IQ, a new intelligence layer designed to give enterprise AI agents deep access to company data, business processes, and live web information.
The platform bundles four distinct context engines, a new agent hosting service, and fresh AI models under one roof. It is now generally available for GitHub Copilot, Microsoft Foundry, and Copilot Studio.
Four flavors of intelligence
Microsoft IQ is not a single product so much as a stack of interconnected data layers, each feeding AI agents a different type of context.
Work IQ pulls signals from Microsoft 365, the productivity suite that already lives inside most large organizations. Calendar patterns, document activity, communication flows: all of it becomes grist for agent reasoning. The Work IQ APIs are expected to launch on June 16, 2026.
Fabric IQ connects agents to structured business data through OneLake, Microsoft’s unified data lake. Instead of an AI agent guessing about quarterly revenue trends, it can query the actual numbers sitting in your analytics infrastructure.
Foundry IQ handles knowledge retrieval across enterprise sources, pulling relevant information from documents, knowledge bases, and internal repositories.
Web IQ is a model-agnostic web passage retrieval API that delivers grounding data from the open internet at roughly 2.5 times faster than Microsoft’s previous technology. Model-agnostic means it is not tied to any single AI model, so developers can plug it into whatever reasoning engine they prefer.
Microsoft is calling the problem it aims to solve the “context gap,” and IQ is its attempt to close it with a unified set of signals that let agents reason, coordinate, and act without requiring bespoke integrations for every data source.
Hosted agents and new AI models
Alongside IQ, Microsoft introduced the Foundry Agent Service, which lets developers deploy hosted, long-running stateful agents. Most AI interactions today are stateless: you send a prompt, you get a response, and the system forgets the conversation ever happened. Stateful agents, by contrast, maintain context across sessions and can execute multi-step tasks over extended periods.
The Foundry Agent Service also includes advanced tracing and optimization features. Tracing is the ability to audit what an agent did and why, a requirement for enterprises that need to explain AI decisions to regulators, compliance teams, or executives.
On the model side, Microsoft previewed its MAI AI models, including a new reasoning model called MAI-Thinking-1.
The agent-first enterprise
Build 2026’s overarching theme was what Microsoft is calling an “agent-first” approach to AI deployment. Security and governance received heavy emphasis throughout the announcements, with Microsoft building governance into the platform layer rather than adding it later.
What this means for investors
The Web IQ component is particularly notable. A fast, model-agnostic retrieval API that any AI model can use to ground its responses in real-time web data could become foundational infrastructure for the agent economy.
None of these announcements included any crypto or token components. Microsoft is making a deliberate bet that enterprise AI adoption will be driven by traditional software infrastructure, not decentralized protocols.