Microsoft’s Build 2026 turns agents into always-on work

Scout always-on – From an always-on assistant called Scout to a new Intelligent Terminal, Microsoft’s Build 2026 leaned hard into AI agents—how they work in the background, how developers run them, and how the company is trying to make them safer, even as it pushes its own AI m

Microsoft kicked off Build 2026 with a keynote led by CEO Satya Nadella and other company leaders—and the theme was clear before the first product details landed: AI isn’t just something you ask questions. It’s starting to do the work while you’re busy.

The announcements ran from new Surface hardware to Microsoft’s latest push to build its own AI models, but the most attention went to the idea of agents that can stay on, operate across apps, and move between devices—if developers can keep them within guardrails.

One of the clearest signals came through Microsoft’s new always-on assistant, Scout. Built on OpenClaw. the open-source AI platform that gained popularity earlier this year. Scout is designed to work with Microsoft 365 apps including Outlook. OneDrive. and Microsoft Teams. Microsoft says Scout is meant to perform tasks in the background. helping businesses with work like organizing calendars. managing expense reporting. writing emails. and more.

Scout is also part of a broader push Microsoft calls “Autopilot” agents, with Microsoft saying each agent will have its own “identity.” For now, Scout is launching in desktop preview for Frontier customers in the US, with Microsoft planning to make it available to more people later.

That “always-on” direction didn’t stop at the assistant. Microsoft also introduced a new Intelligent Terminal aimed at making Windows more developer-friendly. The terminal is designed to give developers context to a preferred AI-powered agent. tying the agent idea directly to the daily workflow of building software.

Microsoft backed that up on the developer side with additions that bring more Linux-like behavior into Windows. It added Coreutils, which Microsoft describes as “Linux-like command-line utilities that run natively” on Windows 11. It’s also launching the ability to create, run, and interact with Linux containers through Windows Subsystem for Linux (WSL).

And for developers who want AI to run locally. Microsoft showed the Surface RTX Spark Dev Box. geared toward running local AI models on a device. The company positions it as a substitute for Qualcomm’s canceled dev kit. The Surface RTX Spark Dev Box is built around Nvidia’s new Arm-based Spark RTX chip and includes 128GB of unified memory. Microsoft says it comes with preinstalled apps including Visual Studio Code and GitHub Copilot. plus a preconfigured version of Windows 11 Pro with dark mode enabled by default. a simplified taskbar. and no widgets.

Microsoft hasn’t released pricing or the full list of specs yet, but it says the device will be available in the US later this year.

As Microsoft pushed agents deeper into tools and workflows. it also tried to address what happens when those agents can access real devices and real data. Speaking of OpenClaw, Microsoft said it is trying to make AI agents safer to run through Microsoft Execution Containers (MXC). With MXC, developers can set guardrails for what AI agents can access on their devices.

Alongside that, Microsoft is launching an OpenClaw companion app. The app is meant to let users set up their own agent or connect to existing ones, which would run in a sandboxed environment.

The keynote also previewed Microsoft’s larger bet on how agents might live across platforms. Microsoft showed Project Solara, described as an Android-based operating system designed to run agents across a variety of devices. Microsoft said it partnered with Qualcomm and MediaTek to develop the system. The idea. as Microsoft presented it. is that the system could eventually work as a companion to a PC or hand off tasks between devices. During the keynote. Microsoft demonstrated sample devices it said Project Solara could run on: a desktop hub and a digital badge.

All of this sat under Microsoft’s continued effort to rely less on outside model providers and more on its own. During Build, Microsoft revealed seven new models, including what it says is its first reasoning model. Called MAI-Thinking-1, the model is said to offer 35 billion active parameters and a 128K context window. Microsoft described MAI-Thinking-1 as designed for “complex multi-step instructions. long-context reasoning and code generation.” Microsoft also announced updates across its models focused on image. voice. and code generation. as well as transcription.

Even as software and AI agents took center stage, Microsoft reserved time for hardware ambition. The company revealed its next-gen quantum computing chip, Majorana 2. Microsoft says the upgraded chip contains qubits that are 1. 000 times more accurate. attributing the performance boost to a new material stack that uses lead and other compounds.

Microsoft also set an aggressive target for that work, saying it could reach its goal of making a practical quantum computer by 2029.

What ties these announcements together is the same direction: agents that can operate on their own. developers who can build and run them with familiar tooling. and guardrails meant to keep them constrained. Scout puts the assistant in the background; the Intelligent Terminal. Coreutils. and WSL container work aim to make Windows the place those agents are actually managed. Project Solara suggests the next step could be spanning devices. And the guardrails—Microsoft Execution Containers and the OpenClaw companion app running agents in a sandbox—signal that Microsoft understands the question people are already asking: not just whether agents can do more. but whether they can do it safely.

Microsoft Build 2026 Satya Nadella Scout assistant OpenClaw Autopilot agents Intelligent Terminal Coreutils Windows Subsystem for Linux Surface RTX Spark Dev Box Project Solara Microsoft Execution Containers MAI-Thinking-1 Majorana 2 quantum chip