At Build 2026, held in the first week of June, Microsoft introduced a new category of AI agent it calls Autopilots: always-on agents that work autonomously on a user's behalf, under their own identity, without needing to be prompted each time. The first product in the category is Microsoft Scout, and it marks a genuine shift in what Microsoft is shipping. Copilot waits to be asked. Scout is designed to notice, decide and act while nobody is watching it.
Coverage of Scout in Australia has been close to non-existent, and some of what circulates internationally mixes confirmed facts with speculation. This guide sticks to what Microsoft and credible technical press have actually published, and is explicit about what remains unconfirmed, including general availability timing and any Australian availability at all.
What Scout is
Scout is an always-on personal agent, delivered as a desktop application for Windows 11 and macOS 12, built on OpenClaw, the open-source agent framework originally created by Peter Steinberger. It integrates with Microsoft 365 and runs continuously rather than per-prompt: managing tasks across email, calendar and reports, and carrying priorities forward over time instead of resetting with each conversation.
Because it inherits OpenClaw's architecture, Scout can perform highly privileged local operations: reading and writing local files, executing shell scripts, applying code patches, launching sub-agents for parallel tasks, and automating browser sessions. It also supports Model Context Protocol (MCP) servers, so its reach can be extended to local resources and third-party tools. That capability profile is exactly why the governance design, covered below, is the most important part of the product.
Work IQ: the grounding layer
Scout's understanding of an organisation comes from Work IQ, Microsoft's intelligence layer that draws on signals from SharePoint, Teams, Outlook, OneDrive and Dataverse and persists an understanding of how individuals and teams actually work: who works with whom, on what content, in what rhythm. Work IQ ships with scoped permissions, policy enforcement and runtime observability, and Microsoft's Build announcements positioned it as production-ready intelligence for agents generally, not just for Scout. Microsoft first announced it was rolling Work IQ into Microsoft 365 Copilot's Wave 3 experiences in March 2026; Build extended it to the agent platform.
The governance design: an agent with its own identity
OpenClaw's power in unmanaged hands has already produced publicised incidents, and security commentary on the open-source project has been blunt. Microsoft's enterprise answer is to wrap Scout in the same identity and data controls that govern people:
- Each Scout instance is assigned its own governed Microsoft Entra identity, so it acts as an attributable principal in the directory rather than under a shared service account.
- Credentials are scoped to individual tasks and redacted from diagnostic logs.
- Microsoft Purview sensitivity labels and data loss prevention policies bind what the agent can touch.
- Highly sensitive operations require human sign-off before execution.
- Policy conformance runs through Agent 365, Microsoft's control plane for agents.
This is the pattern worth internalising even for organisations that never deploy Scout: autonomous agents as first-class directory citizens, with per-task credentials and human approval gates on sensitive actions. It is a preview of how agent governance will look across the Microsoft stack.
How the preview actually works, and what is unconfirmed
Access today is through Microsoft's Frontier preview program, and it is deliberately gated. An organisation must attest to the Frontier terms in the Microsoft 365 admin centre, IT administrators must explicitly push the desktop app to user machines via an Intune policy, and users must hold an active GitHub Copilot Business or Enterprise licence to authenticate. There are two variants in circulation: the Frontier desktop preview, which runs on the user's own hardware and connects to Microsoft Graph and Work IQ, and a separate cloud-based version of Scout in private preview.
What Australian organisations should do while they wait
- 1Treat the prerequisites as the roadmap. Scout's gated rollout assumes an organisation already has admin-centre governance attestation, Intune app deployment and agent-aware identity practices in place. Those can all be built now, and each is useful regardless of Scout.
- 2Sort agent governance before autonomous agents arrive, not after. Scout conforms to Agent 365 policy; an organisation with no agent inventory or policy has nothing for it to conform to. New Agent 365 purchases have required an E5-class base licence since 1 June 2026, which takes lead time to plan.
- 3Decide the human-approval boundary in advance. Which actions in your environment should always require sign-off, whoever or whatever initiates them? That policy question needs an answer before any always-on agent gets credentials, and writing it down costs nothing today.
- 4Pilot the interactive rung first. Organisations still early with Microsoft 365 Copilot and Copilot Studio agents will get more from maturing those than from queueing for an Autopilot preview.
Try it
How ready is your organisation for autonomous agents?
Model what an agentic workforce would look like in your environment, and which governance gaps would bite first.
3 questions · 90 seconds
Which agents would augment your team first?
Pick your industry, team size, and the top three back-office queues currently clogging your team. Frontrow returns the agent recommendations that fit, with what each one would augment.