Microsoft held Microsoft Build 2026 on Tuesday and said it updated its platform to support developing and operating AI agents and applications.

Satya Nadella (사티아 나델라), Microsoft CEO and chairman, said the core of Build 2026 is not a single technology but supporting the creation and scaling of value on a platform. He said Microsoft is expanding the foundation for developers to create greater value through an AI stack that runs from the edge to the cloud.

Microsoft highlighted a multi-model ecosystem spanning local PCs to the cloud so developers can build technology while maintaining choice and control.

The company said key themes at the event can be summarised as building, deploying and optimising agents based on the Microsoft Agent Platform powered by Microsoft IQ; full-stack flexibility and local execution options spanning silicon, operating systems, developer tools and the cloud; and agent systems expanding into science and computing, along with expanded outcomes for researchers and scientists.

Microsoft formally unveiled Microsoft IQ, a new context layer. Microsoft IQ is provided across GitHub Copilot, Microsoft Foundry and Copilot Studio, and supports agents by linking internal enterprise knowledge with external knowledge.

Microsoft IQ consists of 3 core intelligence layers: Work IQ, Fabric IQ and Foundry IQ.

Work IQ is a work intelligence layer that captures work flows and connections across Microsoft 365, organisational systems and external sources. The Work IQ API will be released on the 16th. Fabric IQ provides a foundation for sharing structured business data, and Foundry IQ supports search planning across enterprise knowledge and the live web. Web IQ is a model-agnostic, MCP-native, AI-first web search stack that helps agents use real web information faster.

Microsoft also unveiled Microsoft Scout, a personal agent for work. Scout is built on OpenClaw and Work IQ, and supports proactively handling meeting preparation, schedule conflicts and repetitive tasks by understanding how users work and using tools such as Teams and Outlook.

Microsoft also unveiled the MAI model family, its in-house AI models. MAI-Thinking-1 is Microsoft’s first reasoning model and was trained from scratch on enterprise-grade curated commercial-licence data without a distillation process.

It is a mid-sized model with 35 billion active parameters and a 256K context window, and is designed to deliver high efficiency and performance at a low token cost. It is strong in executing complex multi-step instructions, long-context reasoning and code generation, and is currently available in a private preview on Foundry.

On security and governance, Agent 365 for local agents was unveiled. It extends Microsoft Entra, Microsoft Defender and Microsoft Purview into a single control plane to observe, control and protect agents regardless of hosting location or underlying framework type.

Microsoft also presented an open end-to-end trust stack that can be applied to AI agents regardless of framework. The trust stack is based on ASSERT, an open-source project that handles policy-based safety evaluation, and the Agent Control Specification, which standardises where and how to apply controls during agent behaviour. A new multi-model agentic security system, codenamed MDASH, uses more than 100 agents to infer data flows, business logic and exploit chains, and delivers context-based remediation measures through the Defender portal.