Microsoft has officially pulled back the curtain on Scout, its first Microsoft 365 \"Autopilot\" agent, at Build 2026 on June 2. The always-on, background worker represents a fundamental shift from the familiar Copilot chat paradigm, promising to monitor Outlook, Teams, OneDrive, and SharePoint round the clock, then take governed actions without waiting for a prompt.

Unlike the reactive, conversational Copilot, Scout is designed to operate autonomously—scanning communications, files, and calendars, then executing multi-step workflows based on policies set in Entra ID and Purview. It can flag urgent emails, summarize missed meetings, reorganize SharePoint libraries, or even draft responses, all while respecting granular permission boundaries.

The announcement, delivered during the opening keynote by Chief Product Officer Scott Guthrie, was greeted with a mixture of anticipation and caution. \"Scout isn't a chatbot; it's a digital colleague that knows your context, respects your rules, and never sleeps,\" Guthrie said.

Autopilot: a new class of AI worker

Scout is the first instance of what Microsoft calls Autopilot agents—AI co-workers purpose-built for specific business functions. While Copilot assists in real time, Autopilot agents operate asynchronously. They have access to the Microsoft Graph, can reason across services, and are governed by the same identity and compliance frameworks that IT administrators already manage.

The naming is deliberate. Where Copilot assists a user in control, Autopilot can take the controls when authorized. Microsoft envisions a fleet of these specialized agents: one for sales, one for customer service, one for engineering. Scout is the first general-purpose office agent, targeting the knowledge worker's digital clutter.

How Scout works under the hood

Scout leverages a multi-model architecture built on GPT-5 and a new reasoning engine optimized for enterprise tasks. It maintains a persistent memory of a user's interactions, but that memory is scoped strictly to the individual's own graph data—no cross-tenant data mixing, Microsoft emphasized.

Key capabilities include:

  • Inbox triage: Scout can categorize incoming emails, highlight those requiring action, and draft concise replies based on past conversation style and organizational knowledge.
  • Meeting intelligence: For meetings a user misses, Scout generates a summary, extracts action items, and adds them to Microsoft Planner or To Do.
  • Content organization: It can detect outdated or duplicate files in OneDrive and SharePoint, suggest archival, and restructure folders according to team conventions.
  • Proactive alerts: Scout monitors for anomalies—like a sudden spike in file sharing or an unusual login—and alerts the user or security team via Teams.

All actions are logged in the Microsoft 365 compliance center. Users can set their own preferences via a new \"Autopilot settings\" portal, where they define what Scout may do automatically and what requires explicit approval.

Governance at the core

The critical differentiator for Scout is its deep integration with Microsoft Entra ID (formerly Azure AD) and Microsoft Purview. Before Scout can perform any action, it evaluates the user's role-based access controls, conditional access policies, and data loss prevention rules. If a document is classified as \"Confidential,\" Scout will not include it in a summary unless the recipient is explicitly authorized.

\"We built governance in from day one,\" explained Sarah Bird, Microsoft's Responsible AI lead. \"Scout operates within a sandbox bounded by the same compliance boundaries that govern everything else in your tenant. It's not a separate entity; it inherits your existing security posture.\"

Administrators can set tenant-wide policies to limit Scout's autonomy. For example, an organization might allow Scout to read all emails but only draft replies for certain departments, or prohibit it from accessing Teams channels labeled as \"Legal.\" All policy decisions are auditable, with reports available in Purview.

Scout vs. Copilot: complementary, not competitive

A natural question is how Scout differs from Microsoft 365 Copilot. Copilot remains the user-facing assistant that responds to queries and generates content on demand. Scout, by contrast, is always on, working in the background like a digital executive assistant. The two are designed to work together: a user might ask Copilot \"What did I miss?\" and Copilot could surface the digests Scout already prepared.

Pricing details were not announced, but Microsoft indicated Scout would be available as part of existing Microsoft 365 E5 subscriptions or as an add-on. A public preview is slated for July 2026, with general availability expected in October.

Early community reaction

Although the Build 2026 session detailed Scout's feature set, the Windows community is already buzzing with speculation. Discussion threads on Windows Forum reflect a common split: many users are excited about the prospect of an AI that can cut through information overload, while others fear surveillance creep and loss of control.

One early benchmark—an informal test by a Microsoft MVP—showed Scout reducing the time to process a 200-email inbox from 45 minutes to 9 minutes, with 92% accuracy in categorizing urgent versus non-urgent messages. However, the same tester noted that Scout occasionally misjudged sarcasm in Teams chats, leading to inappropriate action suggestions.

Privacy advocates have also raised concerns. Because Scout needs access to all communications and files to be effective, some users worry that the agent could become a data hoarder. Microsoft has countered that all processing happens within the tenant boundary, no data leaves the Microsoft 365 trust boundary, and personal data is never used to train base models.

Practical considerations for IT admins

Rolling out Scout will require a thorough review of existing governance policies. Microsoft recommends a phased approach: start with a pilot group, define clear opt-in consent, and initially restrict Scout to read-only actions. The company is shipping a set of prebuilt Purview compliance templates specifically for Autopilot agents.

A key decision point is the scope of Scout's memory. By default, it remembers up to 30 days of interaction history. Admins can reduce this to as little as 24 hours for highly sensitive environments. The memory is not shared across users, and there is no centralized store of all user conversations; each instance remains siloed.

Security implications

From a security standpoint, Scout introduces a new attack surface: an autonomous agent with broad read access. Microsoft has hardened the service with just-in-time elevation, where Scout requests temporary permissions for specific actions and releases them immediately. It also uses the same phishing and malware detection engines that protect Exchange Online and SharePoint, so a malicious attachment would be blocked before Scout could process it.

Microsoft's red team has been attacking Scout internally for over six months, and the company is standing up a bug bounty program in August 2026. Details will be published on the Microsoft Security Response Center.

The bigger picture: Autopilot and the future of work

Scout is merely the beginning. During the Build session, leaked slides—quickly confirmed by Microsoft—showed a roadmap for domain-specific Autopilot agents: DealMaker for sales pipelines, CodeWarden for developer code reviews, and SupportPilot for helpdesk ticketing. The common thread is governance; every agent is bound by the same Entra ID and Purview framework.

This aligns with Microsoft's broader vision of \"AI at work\" that respects organizational boundaries. CEO Satya Nadella has repeatedly stated that enterprise AI must be \"secure, compliant, and private.\" Scout represents the first tangible product built from the ground up with that mantra.

Analysts are calling it a response to Google's Gemini Enterprise agents and Salesforce's Einstein Copilot, but with stronger emphasis on IT control. Forrester analyst Mark Bartell noted that \"Scout finally gives Microsoft a plausible story around proactive AI that doesn't sacrifice compliance—something rivals have struggled to articulate.\"

What to expect next

The public preview in July will be limited to English (US) and 1,000 tenants. Microsoft says it will prioritize organizations with advanced Purview configurations, presumably to stress-test the governance model. A broader rollout will follow in September for E5 subscribers, with SMB plans under consideration.

For now, the message from Redmond is clear: AI assistants that only answer questions are yesterday's news. The next wave is agents that act, and Scout is the standard-bearer. Whether enterprise customers embrace a digital colleague that reads their email before they do remains to be seen, but one thing is certain—Microsoft has placed a big bet that governable autonomy is the right answer.

As users prepare for the preview, the advice from admins on Windows Forum is consistent: review your sensitivity labels, tighten your conditional access policies, and get your Purview workload classifications in order. Scout is coming, and it will be only as good as the governance guardrails you set.