Microsoft Tests Always-On OpenClaw-Style Agents for 365 Copilot

Microsoft has confirmed it's experimenting with OpenClaw-style autonomous agents inside 365 Copilot. According to The Information's report, the goal is an "always working" assistant that can complete multi-step tasks over days or weeks without user prompts. Omar Shahine, Microsoft's corporate VP, told the outlet they're "exploring the potential of technologies like OpenClaw in an enterprise context" with beefed-up security controls compared to the open-source original.

The new agents would monitor Outlook inboxes and calendars continuously, then surface daily task lists. Unlike existing cloud-based Copilot Tasks (consumer-focused) and Copilot Cowork (enterprise cloud agent), this system appears designed to run locally on user hardware — a key distinction from Microsoft's current agent lineup that all operate in the cloud.

OpenClaw's rise came with baggage: security researchers documented agents going rogue, prompt injection attacks, and unrestricted access to sensitive data. Microsoft's enterprise customers won't tolerate those risks. The company is engineering role-specific agents (marketing, sales, accounting) with limited permissions to silo access — a direct response to OpenClaw's "all-access" approach that terrified IT departments.

This represents a fundamental shift from Microsoft's cloud-first strategy. Local processing means sensitive corporate data never leaves the device, addressing compliance headaches for financial and healthcare clients. The trade-off: Microsoft must solve how to maintain model performance without cloud compute, particularly for complex multi-day workflows.

Redmond now has three distinct agent products: Copilot Tasks (prosumer cloud), Copilot Cowork (enterprise cloud with Claude integration), and this new OpenClaw-like local agent. Each serves different use cases, but the overlap is confusing. Cowork already handles multi-step workflows using Anthropic's Claude, while Tasks focuses on simpler automation. The local agent appears aimed at power users who need 24/7 autonomy without cloud dependency.

This fragmentation mirrors Microsoft's broader AI strategy — throw multiple approaches at the wall and see what sticks. The company has partnered with both OpenAI and Anthropic, runs its own models, and now experiments with open-source inspired architectures. For customers, it means navigating a maze of overlapping products with unclear differentiation.

OpenClaw's grassroots success forced Microsoft's hand. The open-source project's popularity among developers, particularly its Mac Mini user base, demonstrated pent-up demand for truly autonomous agents. Unlike SaaS tools that require constant connectivity, OpenClaw runs locally, making it attractive for privacy-conscious users and offline workflows.

Microsoft's response shows how open-source innovation drives enterprise features. By commercializing OpenClaw's core concepts (local agents, persistent task execution) with enterprise-grade security, Microsoft validates the open-source approach while attempting to capture its user base. This pattern — open-source proves demand, big tech scales safely — repeats across AI tooling.

Microsoft plans to preview these features during Build conference starting June 2nd. Expect demos showing agents handling multi-day projects — think: monitoring email for client requests, updating CRM systems, scheduling follow-ups without human intervention. The company will likely emphasize enterprise security wins over raw capability.

Key questions for Build: Will these agents require specific hardware configurations? How will licensing work for always-on local processing? And critically, how does Microsoft prevent the "agent gone wild" scenarios that plague OpenClaw users? The answers will determine whether this becomes a must-have enterprise feature or another experimental footnote.

This move pressures cloud-first competitors like Google Workspace and Salesforce. If Microsoft delivers secure local agents, it flips the script on AI's cloud dependency narrative. Hardware partners (particularly those making enterprise-grade desktops and laptops) stand to benefit as "AI-ready" becomes a selling point.

For developers, Microsoft's embrace of local agent architectures could standardize APIs and security patterns currently fragmented across OpenClaw's ecosystem. The real winners might be enterprises who've waited for AI agents that respect data sovereignty while delivering actual autonomy — not just fancy chatbots with extra steps.