Microsoft used Build 2026 to spell out a broad shift in how it wants Windows, GitHub and Azure to support AI. The company’s message was that chatbots are no longer enough. Instead, it is building software for AI agents that can reason, use tools, access data and operate within enterprise security controls.
Chief executive Satya Nadella described the company’s new AI stack as a combination of compute, models, context, tools, runtime and security. In practical terms, that means Microsoft is trying to supply the hardware, software and oversight needed for agents to do real work rather than simply answer prompts.
Among the most notable announcements were updates to the Windows AI API, including Aion 1.0 Instruct and Aion 1.0 Plan. Microsoft said the models are designed for local reasoning, planning, tool use and on-device agent workflows.
The company also introduced the Surface RTX Spark Dev Box, a local development machine built for AI work. Microsoft said the system offers 1 petaflop of AI compute, 128GB of unified memory and support for running models of up to 120 billion parameters locally.
To give agents more context, Microsoft unveiled a set of new IQ-branded layers. Web IQ is meant to bring in fresh web data, Work IQ to provide company context, Foundry IQ to support model-building context and Fabric IQ to supply data context. Microsoft is positioning these pieces as inputs that help agents understand the environment they are operating in.
GitHub also took a central role in the company’s plans. Microsoft showed off a new GitHub Copilot app that it is framing as a desktop experience built for agents. The app is meant to compete more directly with tools such as OpenAI’s Codex and Anthropic’s Claude Code, while offering access to models from OpenAI, Anthropic and Google.
The broader strategy appears to be that GitHub becomes a control center for coding agents, while Windows serves as the local machine where those agents run. Microsoft 365 is expected to provide memory and context for office work, and Azure’s Foundry layer handles deployment.
Microsoft also emphasized the importance of safeguards. The company pointed to Agent 365, MXC, Defender, Entra and Purview as parts of the guardrail system that should help enterprises monitor and control what AI agents are doing.
That focus on governance reflects the company’s view that enterprise adoption will depend not only on capability, but on trust. Agents that can access data and take actions will also need logging, permissions and security oversight.
Microsoft’s announcements come as other AI companies move in a similar direction. OpenAI said its Codex product has surpassed 5 million weekly users and noted that non-developers account for about 20% of usage, growing faster than developer usage.
OpenAI has also been adding role-specific plugins and tools for fields such as sales, analytics, creative production, product design, investing and banking. It recently added Sites, which lets teams share interactive websites or apps inside Codex through a URL.
The common thread across these products is a push to make AI part of everyday workflows rather than a separate destination. Microsoft’s Build presentation suggested that the company believes the winner in the agent era will be the one that owns the work environment, not just the most impressive demo.
For Microsoft, that means turning Windows into an agent-ready operating system, GitHub into a coding-agent hub and Azure into the infrastructure that connects them all.