Anthropic has expanded its AI assistant Claude with new computer-use capabilities, allowing it to carry out tasks directly on a user's macOS device. Announced on Monday, March 23, 2026, the feature enables users to remotely assign tasks—even from a phone—with Claude executing them on a connected computer. This transforms Claude from a conversational assistant into an autonomous agent capable of handling real-world workflows.
The update allows Claude to open applications, browse the internet, and interact with files such as spreadsheets. In a demonstration video, a user asked Claude to export a pitch deck as a PDF and attach it to a calendar invite, which the AI completed without manual input. A key component is the new Dispatch feature within Claude Cowork, letting users "assign a task from your phone, turn your attention to something else, and come back to finished work on your computer."
The rollout signals a broader industry shift toward "agentic AI," where systems operate independently to complete tasks. This race has been significantly driven by the rapid rise of OpenClaw, a free, open-source project created by Peter Steinberger. OpenClaw allows instructions via messaging apps like WhatsApp or Telegram and runs locally on devices, but requires technical setup with API keys and terminal commands. Anthropic is targeting users who found OpenClaw too complicated, offering built-in connections to tools like Google Drive, Gmail, and Slack for a plug-and-play experience.
Safety concerns are a central differentiator. While OpenClaw requests broad system access, Anthropic emphasizes a more controlled approach: turning on Claude's feature requires a single click, and the AI must request user permission before accessing each new application. The company has also built defenses against prompt injection attacks. Anthropic was upfront about limitations, advising users not to let Claude access apps with sensitive data currently and noting the technology is still in an early research preview phase compared to Claude's strengths in coding and text.
The competitive landscape is heating up. OpenClaw's creator, Peter Steinberger, was hired by OpenAI last month to accelerate work on personal AI agents. Nvidia CEO Jensen Huang recently stated OpenClaw represents the "next major evolution after ChatGPT," and the company introduced an enterprise version called NemoClaw. Microsoft and Google are also building similar technology.
Market ripple effects have already been observed. Earlier this year, news of Anthropic's automation capabilities sent Indian IT stocks lower on concerns over displacing software tasks. In China, where neither Anthropic nor OpenAI operates officially, OpenClaw spread widely, leading to reported sell-outs of Mac Mini computers and subsequent instructions from the Chinese government for state-owned companies to stop running it on office machines.
For now, Anthropic's computer-use feature is available as a research preview on macOS for users on the Claude Pro and Claude Max plans.
