Microsoft tests always-on OpenClaw agents for Copilot

Imagine clocking out for the day, closing your laptop, and knowing your AI assistant is still sitting at your desk, digging through unread emails and organizing your morning calendar.

According to a report from The Information, Microsoft is currently testing OpenClaw-style always-on AI agents inside its Microsoft 365 Copilot suite. Microsoft corporate VP Omar Shahine confirmed that the team is actively exploring the potential of technologies like OpenClaw for enterprise use cases.

Taking open-source experiments to corporate cubicles

OpenClaw gained massive traction earlier this year as an open-source framework that lets users spin up autonomous AI agents locally on their own machines. But that independence came with serious security alarms, as local agents with sweeping permissions can easily wander off the rails.

Microsoft claims it can build a significantly safer enterprise-grade version. The goal is to turn Copilot into a proactive assistant that runs 24/7 in the background, handling routine busywork without waiting for a manual prompt.

In daily practice, that means an agent could continuously monitor your Outlook inbox and calendar, surfacing a prioritized list of suggested tasks every morning so you can hit the ground running.

Walling off the departments

Granting an autonomous bot free rein over enterprise servers is an IT nightmare. To counter that, Microsoft is reportedly developing role-specific agents for marketing, sales, and accounting.

By creating narrow, siloed bots, Microsoft aims to restrict permissions strictly to what a specific job requires. An agent built for marketing campaigns will not have access to sensitive corporate ledgers, preventing accidental data leaks across business units.

Microsoft plans to unveil some of these capabilities at its upcoming Build conference starting June 2nd. The move follows pressure from rivals like Anthropic, which integrated Claude into Microsoft 365 last year and introduced Claude Cowork for multi-step projects.

Why it matters

The AI race is rapidly shifting from chat boxes waiting for prompts to background agents executing real work. If Microsoft pulls this off, enterprise workers get genuine automation for repetitive clerical overhead, like triaging chaotic inboxes.

That said, we think enterprise buyers should remain cautious until clear boundaries are proven. Open-source agent frameworks have repeatedly shown that autonomous software makes unpredictable mistakes when left unattended, and an autonomous bot messing up an email draft or misreading an invoice can create real havoc.

Strategically, Microsoft's decision to silo these agents by department is clever. Enterprise IT departments will never sign off on an all-knowing agent with universal access, so role-based guardrails solve a major adoption hurdle. Now Microsoft has to prove that its corporate-grade OpenClaw is actually as safe as it promises.