Microsoft Scout, premier agent Autopilot permanent, built on OpenClaw and a competitor of Cowork

Microsoft has unveiled Scout, a pioneering entry in a new category of agents called Autopilots. These persistent agents operate in the background within Microsoft 365, governed by their own Entra identity. Built on the open-source framework OpenClaw, Scout autonomously coordinates meetings, agendas, and deliverables without user prompts. The agent is currently available in an experimental version for organizations enrolled in the Frontier program.

Scout is part of a familiar family of productivity tools, joining Claude Cowork from Anthropic, which assists users with files and documents, and OpenClaw, an open-source personal assistant that quickly became a highly followed project on GitHub. Modeled after these existing agents, Scout integrates into Microsoft 365, adopting the “Autopilot” nomenclature introduced by Omar Shahine, Vice President of Microsoft Scout, during the Build conference. The technical foundation of Scout is deeply rooted in OpenClaw.

According to Microsoft, Autopilots are defined as always-on agents with their own identity that understand how work is organized across applications and take initiatives without the need for constant user input. Scout connects with Teams, Outlook, OneDrive, and SharePoint, accessing conversations, emails, calendars, and contacts. It extends across the desktop app, browser, local resources, and MCP servers. Users interact with Scout primarily through Teams, where it schedules meetings across time zones, prepares necessary materials, blocks out time on calendars based on identified deadlines, and flags pending decisions that could hinder project progress. Microsoft employees have been using an early version of this desktop tool for several months, with positive feedback on its coordination capabilities and early risk detection.

A Direct Lineage with Cowork and OpenClaw

The journey began in January 2026, when Anthropic launched Cowork, which emerged from Claude Code’s pivot to general productivity tasks. This prototype established the concept of an agent capable of generating tangible work beyond mere conversation. Around the same time, OpenClaw, initially named Clawdbot, saw rapid adoption, surpassing long-standing projects such as React and Linux in GitHub stars. This autonomous personal assistant gained traction faster than traditional vendors could offer solutions.

Microsoft’s response was twofold. In March, the company introduced Copilot Cowork, its version of Anthropic’s concept, integrated into the E7 bundle priced at $99 per user per month. By June, with the debut of Scout, Microsoft shifted its approach, adopting the community framework while layering it with its own identity, compliance, and administrative protocols. This move marked a departure from Microsoft’s previous strategy surrounding Copilot and proprietary MAI models and acknowledged the rapid adoption of open-source tools, which has outpaced traditional product cycles. Similarly, Google followed suit in May with Spark, its autonomous agent for Workspace. In just five months, this permanent personal agent transitioned from a community prototype to a key offering within both major office suites, becoming an everyday touchpoint for users.

An Agent Identity Governed by Entra

The choice of OpenClaw raises potential security concerns, as the framework has a documented history of vulnerabilities and erratic behavior noted earlier this year in a researcher’s messaging. Microsoft addresses these concerns with its governance layer. Each agent operates under a dedicated, governed Entra identity tied to a recognized entity in the directory, avoiding the pitfalls of anonymous shared service accounts. Access is restricted to approved resources, sensitive actions might require human verification, and Microsoft Purview’s data protection policies—complete with sensitivity labels and leak prevention—apply before any writing or sending occurs.

Microsoft also contributes a compliance mechanism back to the OpenClaw community, allowing organizations using the framework (both Microsoft’s version and the community iteration) to verify that their environments meet security requirements, accompanied by an auditable response. Access to Scout requires enrollment in the Frontier program, an Intune policy configuration, and explicit membership confirmation, creating three prerequisites that limit experimentation to organizations already engaged in agent governance. Pricing details remain undisclosed, and it is unclear whether the agent will be included in the Microsoft 365 Copilot subscription, which costs $30 per user per month and has approximately 20 million paying users.

From Passive Data Risk to Active Operational Risk

For organizations already leveraging Microsoft 365 Copilot, Scout amplifies existing data governance challenges rather than introducing new ones, according to Jeff Pollard, Vice President and Principal Analyst at Forrester. The distinction lies in the move from passive risk to active management. “Instead of merely exposing sensitive data to users, Scout can actively work with that data, turning it into an operational risk in daily activities,” Pollard explains, advising security leads to adopt a nuanced approach to securing these tools rather than rejecting them outright.

The arrival of a permanent agent with its own identity aligns with IT departments’ recent initiatives focused on managing non-human identities. It also intersects with cautious adoption recommendations for agent-based AI services published by six agencies in the Five Eyes alliance in May, which advocate for agent inventories, identity management, and action traceability. For European organizations, an agent that continuously accesses emails, calendars, and documents—operated by a provider subject to the Cloud Act—introduces jurisdictional complexities that the Frontier program’s early version allows time to address.

Scroll to Top