Microsoft is testing ways to integrate OpenClaw-like capabilities into 365 Copilot, aiming to make the assistant “run autonomously around the clock” while completing tasks for users, according to a report from The Information as covered by The Verge. The effort is part of Microsoft’s push to expand what an enterprise AI assistant can do—beyond responding to prompts—toward continuous, scheduled work inside Microsoft 365.
OpenClaw as the reference point
OpenClaw is described as an open-source platform that allows users to create AI-powered agents that run locally on a user’s device. The platform rose in popularity earlier this year but has raised a number of serious security concerns.
Microsoft’s corporate vice president Omar Shahine confirmed to The Information that Microsoft is “exploring the potential of technologies like OpenClaw in an enterprise context.” This phrasing indicates Microsoft is evaluating how agent-like behavior could be packaged for enterprise requirements, where security boundaries and permission controls are central.
From chat to always-on task execution
According to The Information, Microsoft’s approach would include capabilities such as monitoring a user’s Outlook inbox and calendar and serving up a list of suggested tasks each day. In other words, the assistant would not only generate text in response to user requests but would also perform ongoing observation of specific Microsoft 365 signals and propose next actions.
This shifts the assistant’s role from interactive tooling to continuous task support: the system would need to interpret events (messages and calendar items) and convert them into task suggestions. The report states that Microsoft is “confident that it can implement ‘safer’ versions of the tool,” though specific engineering approaches are not detailed in the source.
Role-specific agents and permission limits
Beyond an always-on general assistant, Microsoft is exploring OpenClaw-like agents tailored to specific roles, including marketing, sales, and accounting. The stated goal is to “limit the permissions the agent needs” and to “silo” those agents from other parts of a business, according to The Information.
This design direction suggests segmented agent deployments aligned to job functions. In enterprise environments, that kind of partitioning could reduce risk if an agent behaves incorrectly or overreaches. The emphasis on limiting permissions and siloing agents indicates that permission boundaries are a core part of the agent concept Microsoft is testing.
Timing and competitive context
Microsoft aims to show off some of these features during its Build conference, which begins on June 2nd. The Information notes that last year, Anthropic launched integrations with its Claude AI chatbot inside Microsoft 365 services, bringing Claude Cowork to Copilot to help complete “long-running, multi-step tasks.” This comparison places Microsoft’s efforts in the context of competitive offerings in the enterprise AI space.
The Information suggests that bringing OpenClaw-like capabilities into Copilot could help Microsoft address customer needs for multi-step task automation. The most concrete elements of Microsoft’s reported approach are continuous operation (“around the clock”), Microsoft 365 integration (Outlook inbox and calendar monitoring), and permission-focused agent design (role-tailored agents with limited permissions and siloing).
Source: The Verge