Microsoft is testing ways to integrate OpenClaw-like features into its existing Microsoft 365 Copilot, according to TechCrunch reporting (citing confirmation to The Information). The effort targets enterprise customers and is described as coming with better security controls than the open source OpenClaw project. The company is expected to demonstrate the new agent at Microsoft Build in June, per The Verge.
What Microsoft’s Agent Would Do
OpenClaw is a tool that runs locally on a user’s computer and can create agents to perform tasks on the user’s behalf. Microsoft’s reported goal is to integrate similar features into Microsoft 365 Copilot while emphasizing enterprise security and control. TechCrunch notes that if Microsoft produces its own version of a local agent, it would join a set of other agentic tools Microsoft has announced in recent months.
One of the main features Microsoft told The Information is that the agent would be described as a version of 365 Copilot that operates continuously and can take actions at any time. The source frames this as an approach for multistep tasks over long periods, which differs from many chat-based workflows that require a user to prompt each step.
However, the source also makes clear that it is not yet confirmed whether Microsoft’s agent would run locally or would instead adopt some of OpenClaw’s features while running in the cloud. That distinction matters because local execution and cloud execution can imply different constraints around access, auditability, and how enterprise security teams integrate agent behavior into existing policy.
How This Fits With Microsoft’s Existing Agent Lineup
Microsoft’s reported agent work does not appear in isolation. TechCrunch outlines two related products that have been announced or released in the past few months: Copilot Cowork and Copilot Tasks.
Copilot Cowork was announced in March. The source says it is designed to take actions in Microsoft 365 apps, rather than only providing search results or chat in a separate work pane. Cowork is powered by Microsoft’s own “Work IQ” technology, which the source describes as an intelligence layer that personalizes Cowork for the user across Microsoft 365 apps.
For the underlying model, Microsoft has tapped Anthropic’s Claude, after partnering with Anthropic late last year. The source adds that Microsoft added Claude as an option for Cowork. It also notes that OpenClaw can work with multiple models, and that Claude remains the model of choice for many users of the open source project. However, Cowork does not run on local hardware; the source says it runs in the cloud. That distinction provides a baseline for how Microsoft may think about agent behavior: the company has already been moving toward agents that act in-app, but the source indicates Microsoft has not yet confirmed whether the new agent would change the deployment model toward local execution.
Copilot Tasks was introduced in February and released in preview. The source says the marketing materials positioned Copilot Tasks toward prosumers, with tasks spanning from organizing email (a Microsoft 365-like task) to organizing travel and appointments (tasks outside Microsoft’s Office suite). Like Cowork, Copilot Tasks runs in the cloud, according to TechCrunch.
Putting these pieces together, Microsoft’s reported agent concept could represent an effort to converge on a single behavior model: agents that can act (not just answer), operate over multistep workflows, and potentially do so with enterprise-grade controls.
Local Execution and Security Considerations
Security is a central theme in the source. Microsoft’s reported testing is described as being geared toward enterprise customers with better security controls than the open source OpenClaw agent. OpenClaw’s local execution design allows it to run on Windows machines and Mac systems.
Because Microsoft is considering an agent approach, the local-versus-cloud question becomes a practical engineering and operations issue. If Microsoft builds a local agent, it may need to address security concerns while fitting into enterprise IT environments. If Microsoft keeps execution in the cloud, it may instead focus on controls around action authorization, auditing, and policy enforcement.
What to Watch at Build in June
According to the source, Microsoft is expected to show off the new agent at Microsoft Build in June, as reported by The Verge. That timing suggests Microsoft wants to position the work alongside its broader Copilot and agent roadmap.
For observers, the key unresolved technical questions are explicit in the reporting: whether the agent runs locally, and whether it represents a distinct product or primarily a Copilot extension that adopts some agent behaviors. The source also frames the agent’s distinguishing capability as operating continuously and taking actions at any time, with multistep tasks over long periods—features that, if demonstrated, would further define the boundary between chat assistance and task-executing automation.
Microsoft has already built cloud-based agents that can act in Microsoft 365 apps (Copilot Cowork) and complete tasks spanning Microsoft 365 and beyond (Copilot Tasks). A next step that adds continuous operation and enterprise security controls could indicate an effort to standardize agent execution patterns across Microsoft’s Copilot suite. At the same time, the source’s lack of confirmation on local deployment means it remains possible Microsoft will choose a cloud-based approach.
Source: TechCrunch