Microsoft is testing an OpenClaw-like agent for Microsoft 365 Copilot, part of a broader effort to turn Copilot from a prompt-based assistant into something closer to an always-on workplace operator. Reporting from The Verge and TechCrunch says the company is experimenting with a version of Copilot that could keep working in the background, handling multi-step tasks over time rather than waiting for a user to ask for each action individually.
That shift would mark a meaningful change in how Microsoft positions Copilot. Until now, much of the product has centered on helping users draft, summarize, and respond inside Microsoft 365 apps. An always-working agent would push it further into orchestration, where the software is expected to monitor context, carry out sequences of tasks, and keep progress moving with less manual prompting.
What OpenClaw is, and why Microsoft is paying attention
OpenClaw is an open-source personal AI assistant that promotes itself as “the AI that actually does things,” with capabilities such as clearing inboxes, sending emails, managing calendars, and checking users in for flights. Its appeal has come from giving power users a way to run more autonomous agents outside the tighter guardrails of mainstream enterprise software.
That openness is also the reason large companies view it cautiously. Tools like OpenClaw can be flexible and powerful, but broad autonomy creates obvious concerns around permissions, oversight, and security. Microsoft’s apparent goal is to capture the appeal of that model while placing it inside Microsoft 365’s enterprise controls, where identity, permissions, governance, and compliance policies are already central to the product pitch.
Part of a much larger Copilot agent strategy
The OpenClaw-style experiment is not happening in isolation. Microsoft has already been moving Copilot toward more action-oriented behavior across Microsoft 365. In March, the company introduced Copilot Cowork, which it described as a system that turns intent into action across Microsoft 365, automating tasks and coordinating workflows while keeping users in control. Microsoft says Cowork can ground itself in emails, meetings, messages, files, and data, then keep tasks moving in the background with checkpoints and approvals.
Microsoft has also been building out a wider agent ecosystem around Copilot. Its February 2026 update for Microsoft 365 Copilot highlighted agents, integrations, and new access points across products such as Teams, Outlook, OneDrive, and PowerPoint, showing that the company is steadily expanding Copilot from a chat layer into an agent platform.
That makes the OpenClaw-style project less surprising than it first appears. Rather than representing a sudden shift, it looks like the next logical step in Microsoft’s campaign to make Copilot more autonomous, more persistent, and more embedded across day-to-day work.
The biggest unanswered question is where the agent will run
One of the most important details is still unclear: whether Microsoft’s new agent will run locally in a way that resembles OpenClaw more closely, or whether it will remain cloud-based like Cowork. OpenClaw’s appeal has partly come from the fact that users can run it in their own environments and connect it to different tools and models. Microsoft, by contrast, has emphasized that Copilot Cowork runs inside a protected, sandboxed cloud environment designed for enterprise-scale security and governance.
That distinction matters because local and cloud agents solve different problems. Local execution can offer more direct control over apps and workflows on a machine, but it also creates more risk and complexity for enterprise IT teams. Cloud-based execution fits Microsoft’s existing enterprise model more naturally, especially when auditability, policy enforcement, and centralized controls are part of the selling point. At this stage, the company has not publicly clarified which side of that divide this new OpenClaw-style effort will land on.
Why this matters for Microsoft’s AI ambitions
Microsoft’s interest in an OpenClaw-like system shows how quickly the AI competition is moving from chat to action. The next battleground is not just which assistant can answer questions best, but which one can reliably do work over time while staying inside the rules enterprises require. That is exactly where Microsoft believes it has an advantage: it already controls the productivity stack, the identity layer, and the enterprise governance model those customers depend on.
The company is expected to discuss the new OpenClaw-style agent, or a closely related upgrade to its existing agent lineup, at its Build developer conference on June 2, according to The Verge’s reporting. Until then, Microsoft is presenting the work as experimental. Even so, the direction is clear. Copilot is being pushed beyond a passive assistant role and toward something much more continuous: software that does not just respond to work, but increasingly stays in the flow of work itself.
