Microsoft Is Building Another OpenClaw-Style AI Agent—Here’s What It Means

The Open-Source Agent That Made Microsoft Move

When Peter Steinberger released OpenClaw as an open-source project, it drew hundreds of thousands of GitHub stars and forks within weeks. Developers and knowledge workers built workflows on top of it to clear inboxes, send messages, manage calendars, browse the web, book travel, run code reviews, and execute tasks across multiple applications simultaneously. Users control it through messaging apps like Telegram and WhatsApp, issuing instructions in plain language while the agent works in the background and notifies when jobs are done.

The Mac Mini became the unexpected hardware of choice for OpenClaw users, its combination of low cost, low power draw, and adequate performance making it the default dedicated machine for running the agent around the clock. Sales of the small desktop rose noticeably as adoption spread.

What OpenClaw demonstrated was a specific kind of value proposition: a persistent, always-on agent with access to your real accounts, your real files, and your real tools, capable of completing complex multistep tasks over extended periods without constant prompting. No major enterprise software company had shipped that. Most AI assistants still operated on the prompt-response model, completing one task at a time when asked.

Now Microsoft is building its answer. The company confirmed to The Information that it is testing OpenClaw-like features inside Microsoft 365 Copilot, targeted at enterprise customers. The project is led by a small internal team under Microsoft corporate vice president Omar Shahine, informally known within the company as "Ocean 11." Shahine announced the role publicly on March 31, 2026, posting on X that his goal is to "help usher in a new generation of workplace proactive assistants, ones that lighten your load by taking on tasks end-to-end." Microsoft CEO Satya Nadella has made revamping 365 Copilot a stated top priority.

What Microsoft Says It Is Building

The core feature Microsoft described to The Information is an always-on version of 365 Copilot, one that works continuously and can take actions at any time rather than waiting for a user to open the application and issue a command. The idea is an agent that can complete multistep tasks over long periods, proactively managing work rather than reactively responding to individual prompts.

The framing is clearly OpenClaw-inspired: background operation, persistent context, task execution across tools. The differentiation Microsoft is staking out is the security and governance layer. The new agent is positioned for enterprise customers specifically, with access controls, identity management, compliance boundaries, and auditability built in from the start.

What remains unconfirmed is whether this agent would run locally on the user's hardware, as OpenClaw does, or in the cloud. Microsoft has not disclosed that detail. The distinction matters significantly for enterprise deployment models, latency characteristics, and the security tradeoffs involved.

Microsoft is expected to preview the agent, or an upgraded version of one of its existing agentic tools, at Microsoft Build 2026, scheduled for June in San Francisco.

Why the Security Problem Is Central

Microsoft published its own security blog post on running OpenClaw safely in February 2026, and the document is notably blunt. The post states that OpenClaw "includes limited built-in security controls," that the runtime "can ingest untrusted text, download and execute skills from external sources, and perform actions using the credentials assigned to it," and that running it on a standard enterprise workstation is not appropriate.

The Microsoft Security team identified three specific risks that materialize quickly in unguarded deployments. Credentials and accessible data can be exposed or exfiltrated. The agent's persistent memory can be modified by attacker-supplied instructions that arrive through untrusted input channels. The host environment can be compromised if the agent is induced to download and execute malicious code through its skills ecosystem.

The post recommended that if organizations must evaluate OpenClaw, they should do so only in a fully isolated environment using dedicated, non-privileged credentials with no access to sensitive data. That is a significant limitation for any enterprise that wants to actually use the agent for real work.

This is the gap Microsoft is positioning itself to fill. An OpenClaw-like agent that operates with the same persistent, always-on capability but runs within Microsoft's existing enterprise identity management, access controls, and compliance infrastructure removes the deployment friction that has kept OpenClaw primarily in individual developer and enthusiast workflows rather than corporate IT environments.

How It Fits Into Microsoft's Existing Agent Stack

The new agent is not arriving in isolation. Microsoft has been building toward this category across several products simultaneously, and the new effort slots into a growing stack.

In February 2026, Microsoft released Copilot Tasks in preview. The product was described as a to-do list that executes itself, with Copilot using its own browser and compute to browse the web, manage schedules, send emails, coordinate across applications, and take actions in the world on the user's behalf. Copilot Tasks runs in the cloud and was positioned initially at prosumers and individual users rather than enterprise teams.

In March 2026, Microsoft announced Copilot Cowork, which targets enterprise users more directly and is designed to take actions inside Microsoft 365 applications rather than just provide chat responses or search results. Cowork is powered in part by Anthropic's Claude, following Microsoft's partnership with Anthropic that deepened in late 2025. Cowork also runs in the cloud, built on top of Work IQ, Microsoft's intelligence layer that tries to personalize Copilot's behavior using context from across a user's Microsoft 365 environment. Cowork became available via Microsoft's Frontier program on March 30, 2026.

The new OpenClaw-like agent represents a third direction within this stack: more autonomous, more persistent, and with a deployment model that has not yet been finalized. Whether it will eventually consolidate the Cowork and Tasks functionality under a single interface, or operate as a distinct product, has not been disclosed.

It is worth noting that OpenClaw itself already works with multiple models, though Claude has become the most popular choice among its users. Microsoft's existing Cowork product uses Claude. If the new agent follows a similar multi-model approach or defaults to Claude for reasoning and execution tasks, the Anthropic partnership becomes increasingly central to Microsoft's overall agentic strategy.

The Broader Competitive Context

Microsoft is not moving in isolation. The enterprise market for persistent AI agents has attracted attention across the technology industry. NVIDIA built NemoClaw, an enterprise security stack layered on top of OpenClaw, with Adobe, IBM's Red Hat, and Box among the organizations that have expressed interest in the approach. Salesforce has acknowledged parallels between OpenClaw's architecture and its own development roadmap for autonomous agents.

OpenAI has Operator, a web automation agent available on its Pro plan, though it operates only within the browser and does not have access to the full desktop environment. Anthropic's Claude Cowork, which Microsoft has already integrated, brings agentic task execution to local files and desktop applications.

What makes the Microsoft move particularly significant is scale. Microsoft 365 has hundreds of millions of enterprise users across organizations of every size. If Microsoft ships an always-on, enterprise-grade agent inside Copilot, it bypasses the adoption barrier that OpenClaw faces with IT teams wary of its security posture. An agent that runs within the boundaries every IT administrator already manages does not require a separate security evaluation, a dedicated isolated VM, or non-privileged credentials. It inherits the trust model the organization already has.

That is the real competitive advantage Microsoft is building toward. Not a better agent in isolation, but an agent that enterprise IT teams will actually allow to run on production systems.

What to Watch For at Build 2026

The preview at Microsoft Build in June will be the first concrete look at how Microsoft has translated the OpenClaw concept into an enterprise product. The details that matter most are deployment model, whether local or cloud-based, the permission and approval model for sensitive actions, how it integrates with or extends Cowork and Copilot Tasks, and which AI models power it.

Microsoft's track record with Copilot has been uneven. Early versions of Copilot in Microsoft 365 drew criticism for failing to deliver the productivity gains promised in marketing. Cowork and the new agent represent a meaningful shift in ambition toward genuine task execution rather than conversational assistance. Whether the execution matches the vision is what Build will begin to answer.

If you are building enterprise software products, AI-integrated workflows, or evaluating how agentic AI fits into your organization's technology stack, please reach out to MonkDA. We work with development and product teams building the next generation of AI-powered enterprise tools.