The Assistant That Doesn’t Wait
Every AI tool most of us have used follows the same basic interaction model: you ask, it answers. You open it, it activates. You close it, it stops. The tool is always waiting for you to start the conversation.
That model is intuitive, controllable, and comfortable. It’s also about to become the old way of doing things.
At Microsoft Build 2026, Satya Nadella introduced a new category of Copilot agent called Autopilots — always-on, autonomous agents that act on your behalf without requiring you to initiate anything. The first one, Microsoft Scout, is available in preview today. And whether you lead a team, manage a Copilot deployment, or just live in your inbox, understanding what Autopilots are — and what guardrails exist around them — is the most practically important thing to come out of Build 2026.
What Makes an Autopilot Different
The word “agent” has been used loosely across the Microsoft AI portfolio for the past year. Copilot Studio agents, GitHub Copilot agents, Foundry agents — the term has started to lose meaning. Autopilots are something specific, and the distinction matters.
A standard Copilot agent is reactive. It runs when triggered — by a user message, a form submission, a webhook, or a scheduled job. When the trigger fires, the agent works. When there’s no trigger, nothing happens.
An Autopilot is proactive. It has its own identity in your Microsoft 365 environment, its own persistent runtime, and its own continuous awareness of what’s happening across your workplace signals. It doesn’t need you to ask it something. It’s already watching, and it surfaces things it determines you need to know.
This is not a trivial difference. It changes the governance model, the trust model, and the change management conversation entirely.
Meet Microsoft Scout
Scout is the first Autopilot and the clearest illustration of what this category is designed to do. Built on the OpenClaw agent runtime, integrated with Microsoft 365, and available as a native desktop application on Windows 11 and macOS 12 and later, Scout monitors two primary streams: your inbox and your Teams sessions.
What it does with that monitoring is important to understand precisely. Scout is not reading your email and acting on it autonomously. In the current preview, Scout observes, surfaces, and suggests. It identifies action items embedded in email threads and meeting conversations — the commitments you made, the follow-ups you promised, the decisions that are waiting on you — and brings them forward in a consolidated, prioritized view.
The demo at Build showed Scout surfacing a “you owe someone a response on this” notification from a thread that had been buried for two days, and a “this decision is blocking three people” flag from a Teams conversation the user hadn’t re-opened. Neither action was taken automatically. Scout flagged them. The human decided what to do.
Technical note: Scout operates with its own M365 identity — distinct from your user account. Its access is governed by admin-controlled permission scopes, and all actions it takes are logged and auditable. It is not acting as you; it is acting on your behalf with a defined, inspectable scope.
The Trust Conversation You Need to Have
There will be a reaction when you introduce Scout — or any Autopilot — to your organization. It will go something like this: “So Copilot is reading all my emails now?”
The honest answer is: yes, with controls, with a defined scope, with audit trails, and with admin-governed permissions. That answer is accurate, but it lands differently depending on how it’s framed. The organizations that navigate this well are the ones that get ahead of the conversation rather than reacting to it.
A few things are worth being clear about when communicating Scout’s rollout to your workforce.
- Scout is opt-in at the user level in the current preview. Users can enable or disable it. It does not activate for everyone the moment it’s deployed to a tenant.
- The permissions Scout holds are explicitly scoped. It does not have access to your entire M365 environment by default. Admins control what it can see and what it cannot.
- Everything Scout does is logged. If a user or an auditor wants to review Scout’s activity, that log exists and is accessible through standard M365 compliance tooling.
- Scout does not take actions autonomously in the current preview. The observe-and-surface model is intentional. Microsoft is being deliberate about where the human-in-the-loop boundary sits at launch.
Framing Autopilots as a controllable productivity tool — not a surveillance system — is a change management responsibility, not just a technical one. Copilot champions and IT leaders who proactively communicate the permission model, the opt-in design, and the audit capability will have a far easier rollout than those who let the rumor mill fill the gap.
What Autopilots Signal About Microsoft’s Direction
Scout is one agent in what Microsoft has explicitly positioned as a new product category. The Autopilot label is intentional — it signals a roadmap, not just a feature release. The analogy to autopilot systems in aviation is apt: the automation handles the routine, the human stays in command, and the level of automation increases incrementally as trust is established.
The logical expansion areas aren’t hard to see. Calendar management — scheduling, rescheduling, protecting focus time — is an obvious next territory. Project tracking, where an Autopilot monitors milestone status across Teams and Planner and flags slipping deadlines without being asked, is another. Procurement and vendor management workflows are a natural fit for organizations that already have Fabric IQ enabled and structured contract data accessible.
Microsoft also signaled that third-party ISVs will be able to build Autopilots on top of Foundry. This means the Autopilot category is not just Microsoft-built agents — it’s an extensibility surface. Expect your ISV ecosystem to start producing Autopilot-category agents over the next 12 to 18 months. Governance policies you put in place now will apply to those agents when they arrive.
What to Do With This Right Now
For most organizations, the immediate priority is preparation, not deployment. Scout is in preview through the Frontier Program, which is the right place to start — get your Copilot champions enrolled, run a structured pilot with clear success metrics, and build your communication plan before broad rollout.
- Enroll in the Frontier Program preview. Scout is available now for organizations in the program. If you’re not enrolled, apply. The feedback loop from early adopters directly influences the GA feature set.
- Define your governance position on Autopilots before users ask. What is your organization’s policy on always-on agents? What data can they access? Who approves new Autopilot deployments? Having a written policy position before the questions arise puts you in a much stronger posture.
- Include Autopilot scope in your Copilot Credits planning. Always-on agents generate consumption differently than user-initiated interactions. Factor this into your Credits budget now rather than after the first billing cycle.
- Prepare a champion-led communication plan. Don’t let the “Copilot is reading my email” reaction happen without a prepared response. Brief your champions, give them accurate talking points, and make the opt-in and audit capability visible and easy to find.
The Shift That’s Actually Happening
Microsoft Scout and the Autopilot category represent something more significant than a new product announcement. They represent a shift in the fundamental relationship between knowledge workers and AI tools.
The first wave of Copilot was about augmenting your work — doing what you do, faster. The Autopilot category is about delegating work — things that were happening (or more often, not happening) because of time, attention, and cognitive load constraints.
The organizations that will get the most from this shift are not necessarily the ones with the most sophisticated technical deployments. They’re the ones that have done the change management work to build genuine trust in AI-assisted workflows — the ones where employees understand what the agent can and can’t do, where governance is visible and credible, and where the human stays confidently in command.
Scout is the starting point. The category it opens is where the next several years of workplace AI are heading. Getting your governance, your communication, and your champion network ready now is the most valuable thing you can do before it arrives at scale.