On 2 June 2026, Microsoft announced a new type of AI agent at Build: the Autopilot. Microsoft Scout is the first product in that category — always on, with its own identity, and capable of acting autonomously across Teams, Outlook and SharePoint. What changes for IT administrators and how do you prepare your environment?
On 2 June 2026, Microsoft unveiled a new type of AI agent at its Build conference: the Autopilot. Where regular Copilot functionality works reactively — you type a question, you get an answer — an Autopilot agent works proactively. It is always active in the background, understands how your work flows, and takes actions without you having to issue a command every time.
The first product in that category is Microsoft Scout. Scout is integrated across Teams, Outlook, OneDrive and SharePoint, has its own governed Entra identity, and operates entirely within the compliance frameworks of your Microsoft 365 tenant. For IT administrators, Scout introduces a new type of entity into the environment: an agent that acts autonomously but falls under the same governance rules as a human user.
The distinction seems subtle but has significant consequences. A regular Copilot task — a meeting summary, a draft email, an answer to a question — is something you initiate yourself by typing a prompt. The interaction is singular: you ask, the model responds, you evaluate the result.
An Autopilot agent works differently. Scout continuously analyses your calendar, email and project context, detects patterns and risks, and takes independent action. That might mean scheduling a meeting based on an identified decision gap, generating a briefing document for tomorrow's meeting, or flagging an impending deadline conflict before you notice it yourself.
The result is that you no longer drive an agent — the agent works on your behalf. That makes the governance requirements fundamentally different from those for a standard Copilot licence.
Scout works across five concrete capabilities. First, calendar optimisation: Scout analyses your calendar, detects duplicates, blockers and meetings with no clear purpose, and proposes adjustments or executes them directly if you have granted that permission.
Second, risk identification: Scout monitors active projects and flags stalled decisions — decisions that have been open too long and are putting a delivery at risk. It can send a summary or draft an escalation email for your approval.
Third, material preparation: for every meeting in your calendar, Scout can automatically generate a briefing document based on email history, linked SharePoint documents and previous meeting notes. You open the meeting with context already prepared.
Fourth, cross-agent coordination: Scout can orchestrate other Copilot agents. Think of an Outlook agent for email, an Excel agent for data processing and a Teams agent for creating a channel — all bundled into one complex task that Scout manages.
Fifth, personal organisation: daily briefings, task overviews and priority lists are assembled from all available context. Scout acts as a digital chief of staff that filters the information flow before it reaches your attention.
Every Autopilot agent operates under its own governed Entra identity. Scout is therefore not a shared service account but an entity with its own directory object, its own RBAC roles and its own audit trail. This makes actions attributable: if Scout sends an email or modifies a document, it is clear who or what authorised that action.
The credentials behind that identity are end-to-end protected: scoped to the task being performed at that moment, redacted from log files, and managed with the same security standard that applies to Microsoft's own services. No permanent, broad access rights are granted.
Microsoft Purview data policies are enforced at the moment of action, before anything is sent or stored. Sensitivity labels and DLP rules therefore apply not only to human users but also to actions carried out by Scout. If a document is labelled Highly Confidential, Scout cannot forward it to an external party, even if the user instructs it to do so.
For IT administrators, the introduction of Autopilot agents means that existing governance frameworks for human users now also apply to automated entities. That requires a number of adjustments.
First, the question of which actions Scout may take without human approval. Microsoft has built an approval model into the product: for sensitive operations — such as sending an email to an external party or modifying a document with a sensitivity label — explicit user approval is always required. IT administrators can use Intune policies to define which action categories are blocked, run in audit mode, or are automatically permitted.
Second, visibility: Scout actions are visible in the Microsoft Purview audit log and in the Defender portals. Administrators can run advanced hunting queries on agent activity and consult the exposure graph to see how Scout connects to systems and data across the network.
Third, containment policies: just as Conditional Access policies determine which users can access from which locations, organisations can now define policies for what agents may and may not reach. Anyone deploying agents in an environment with strict data residency requirements — such as sectors governed by NIS2 or financial services regulation — must explicitly configure this layer.
Microsoft Scout launched on 2 June 2026 in private preview for Frontier organisations. Frontier is an early-adopter programme in which participants gain access to experimental Microsoft 365 functionality in exchange for extensive feedback and technical engagement.
Current participation requirements are: an active Microsoft 365 Copilot licence, a GitHub Copilot licence, enrolment in the Frontier programme, and signing an opt-in declaration confirming the experimental status. Intune configuration is required to set policies for Scout execution.
Pricing for broader availability has not yet been communicated. Microsoft has indicated it will evaluate the Frontier preview before publishing a general availability timeline. For most organisations that are not currently in Frontier, Scout is not yet directly deployable, but the governance implications are already relevant: it is wise to prepare policies and frameworks before the GA date approaches.
Four concrete steps every IT professional can take in the coming weeks, regardless of whether they already have access to Scout or not.
First, audit your current Purview configuration. Are sensitivity labels consistently applied to documents and email? Are DLP rules active and aligned with the sensitive data in your environment? Agents like Scout operate strictly within those frameworks, so an organisation with weak label configuration is indirectly also giving an agent more freedom than is prudent.
Second, ensure your Entra environment is ready for non-human identities. That means understanding the RBAC roles that apply to service accounts and agents, having audit logs in order, and defining a procedure for creating, managing and revoking agent identities.
Third, actively follow the Frontier programme and its accompanying admin documentation. Microsoft publishes technical guidance as new Autopilot features enter preview. Organisations that absorb this documentation quickly can make well-considered participation decisions.
Fourth, involve your Copilot ambassadors or power users early. Autopilot agents like Scout require a different mindset from a prompt-based tool. Employees who are used to driving actions themselves must learn to trust — while remaining critical of — what an agent decides autonomously. That requires internal guidance and clear agreements on which categories of decision always require human approval.
Microsoft Scout is the first concrete realisation of what autonomous AI agents mean in the workplace. It is no longer a distant future but a product in preview with a clear rollout direction. Organisations that get their governance foundations in order now will be positioned to adopt it safely and in a controlled way when general availability arrives. Want to prepare your Microsoft 365 environment for autonomous agents, or have your Purview and Entra configuration reviewed? Contact Zarioh.
Zarioh Digital Solutions
IT specialists from Utrecht, the Netherlands. We help businesses with Microsoft 365, AI agents, hosting and telephony — and share what we learn in practice. Follow us on LinkedIn
