Microsoft dropped a bombshell at Build 2026 with Scout, a personal AI agent that doesn’t wait for your command. Scout runs continuously across Microsoft 365, ingesting your email, files, chats, and calendar to do work before you even ask. It’s the company’s most aggressive push yet to fuse generative AI with the operating system and productivity suite, and it fundamentally changes the relationship between knowledge workers and their tools.

Scout is not a chatbot you summon. It’s an always-on agent built on Microsoft’s new OpenClaw framework. That means it lives inside Teams, Outlook, OneDrive, SharePoint, and even reaches across Windows, macOS, and the web. Its promise: anticipate what you need, draft responses, schedule meetings, surface documents, and take actions on your behalf, all while you focus on higher-order tasks.

Always-on means always working

The defining characteristic of Scout is its persistence. Traditional AI assistants require a prompt; Scout watches signals in real time. If a colleague sends a Teams message asking for the latest quarterly deck, Scout can pull the file from OneDrive, check it against the request, and offer to share it—without you touching the mouse. In Outlook, it might draft a reply to a recurring vendor email based on your past responses and current project status.

That continuous awareness raises obvious privacy and governance questions, and Microsoft addressed them head-on at Build. Scout operates within a new “agent governance” framework that lets administrators set granular policies. IT can define which data sources the agent can access, which actions it can take autonomously, and when it must escalate for human approval. All actions are logged in a unified audit trail, giving compliance officers a clear view of autonomous decision-making.

OpenClaw: The engine under the hood

Scout isn’t a standalone product; it’s the first major application of OpenClaw, Microsoft’s new agentic AI framework. OpenClaw abstracts the underlying large language model so that Scout can use the best model for a given task—GPT-6 for nuanced communication, smaller special-purpose models for classification or extraction. The framework also handles cross-application context, letting Scout maintain a unified memory of user activity across 365 endpoints.

Because OpenClaw is designed to be extensible, third-party developers can build agents that integrate with Scout. Imagine a CRM that plugs its own agent into the same memory fabric, allowing Scout to pull customer history into an Outlook draft automatically. Microsoft sees Scout as the user-facing front end for a broader agent ecosystem, and Build sessions detailed APIs that will let enterprises craft custom agents that inherit the always-on capability.

Productivity without the prompt

Scout’s real-world impact could be profound. Early demos showed it drafting meeting agendas by analyzing ongoing chat threads and flagging action items from previous meetings. It can proactively surface files that colleagues are waiting on, or highlight email chains that need attention based on sentiment and urgency. In one Build keynote scenario, Scout noticed a scheduling conflict, proposed an alternative time, and drafted the reschedule email—all while the user was presenting slides.

This shift from reactive to proactive AI marks a sea change in how Microsoft envisions work. Instead of workers managing applications, the agent manages applications on behalf of the worker. It’s a vision that’s been promised for decades, but the combination of large language models and OpenClaw’s orchestration layer makes it technically feasible for the first time at scale.

Cross-platform, cross-app, cross-device

Scout isn’t confined to Windows. Microsoft made clear that the agent works on macOS and through a progressive web app, ensuring that the growing share of Mac-using Office customers aren’t left behind. On Windows, Scout integrates deeply with the OS—it can tap into notifications, file system events, and OS-level search to build context. On the web, it runs as a browser extension or service worker, so even Chromebook users get a consistent experience.

This ubiquity is critical for an agent that relies on a complete picture of your work. If Scout only saw Teams but not Outlook, its contextual understanding would be fragmented. By spanning all 365 surfaces and the OS itself, it constructs a rich graph of relationships, projects, and deadlines. Microsoft calls this the “semantic workspace,” and it’s the secret sauce that lets Scout make intelligent suggestions.

Governance and the trust problem

Always-on agents court controversy, and Microsoft knows it. At Build, the company spent considerable time detailing governance features. Agent policies are scoped at the tenant, user, and application level. A policy might say: “Scout can read email and Teams, but cannot send messages without confirmation.” Or: “Autonomous actions are allowed only for files tagged with project ‘Alpha’.”

Every action Scout takes is logged with a timestamp, input context, model reasoning trace, and outcome. This audit trail is designed to satisfy regulators in heavily scrutinized industries like finance and healthcare. Microsoft also introduced a “trust index” that scores each action based on confidence level and potential impact, with high-impact decisions requiring human sign-off.

Privacy is handled through on-device processing where possible. Sensitive data can be indexed locally on the user’s machine, with only anonymized vectors sent to the cloud for inference. Enterprises can also deploy OpenClaw on Azure private instances to keep data within their own subscription. These measures aim to defuse fears that an always-on agent becomes a surveillance tool.

Competition heats up

Microsoft isn’t alone in the always-on agent race. Google’s Project Mariner, also announced this year, takes a similar approach within Google Workspace, while Apple’s revamped Siri tries to span Mac and iOS. But Microsoft’s advantage is its enterprise install base. With over 400 million Office 365 commercial seats, Scout has a built-in audience that no competitor can match overnight.

The betting at Build is that Scout will become a primary differentiator for Microsoft 365 licenses. The agent will likely be included in E5 or as a premium add-on; pricing wasn’t confirmed, but executives hinted at a consumption-based model for heavy autonomous actions.

What early adopters say

Pre-release testers in the Microsoft 365 Insiders program have already shared mixed impressions. On forums, some praise Scout’s ability to summarize complex email threads and prepare meeting briefs. One tester noted: “I got a draft reply to a client email that I didn’t even remember I owed. Scout had been tracking the thread for two weeks and nudged me. That alone saved a deal.”

Other testers flagged over-eagerness. Scout sometimes suggests actions based on incomplete context, such as proposing a file share when the document hadn’t been final-reviewed. The “nag factor” surfaced in multiple discussions, with users wanting finer-grained control over when Scout surfaces notifications. Microsoft acknowledged these pain points and promised iterative tuning through the public preview period starting July 2026.

The road ahead

Scout enters public preview in mid-2026, with general availability slated for early 2027. That timeline gives Microsoft months to refine the agent based on Insider feedback and to expand OpenClaw’s third-party ecosystem. By the time Scout reaches GA, Microsoft expects dozens of line-of-business agents to be available, from SAP payroll bots to Salesforce deal-closing assistants.

For Windows enthusiasts, Scout signals a deeper OS-level integration future. Imagine an agent that not only manages your Office docs but also adjusts system settings, runs PowerShell scripts, or troubleshoots errors. Microsoft teased that Windows 12 will treat agents as first-class citizens, with native APIs that let Scout and similar tools tap hardware acceleration for on-device AI.

Scout isn’t just a feature; it’s a new computing paradigm. It asks workers to trust an algorithm with their daily decisions. At Build 2026, Microsoft made a compelling case that the technology is ready, the governance is robust, and the productivity gains are real. The next year will determine whether that case holds up under real-world use.