In a move reflecting the rapid advancement of generative artificial intelligence technologies, Anthropic has announced the launch of new and advanced capabilities that allow its intelligent model, “Claude,” to directly control Mac computers. This radical update shifts AI interaction from a mere text-chat tool into an executive agent capable of directing, clicking, and browsing various applications in the exact same way a human user interacts. The ability of machines to mimic human behaviors in navigating operating systems marks a qualitative leap in the trajectory of technology, where the role of software is no longer limited to providing text answers or generating code, but extends to the actual execution of both routine and complex tasks alike.
Beginning March 23, this innovative update will be available as a research preview for subscribers of the “Claude Pro” and “Claude Max” plans, bringing computer-use features to popular work tools such as “Claude CoWork” and “Claude Code” on the macOS operating system.
A radical shift from a chat tool to an active desktop agent
In a video showcasing the new feature, Anthropic demonstrated Claude’s superior ability to handle a wide range of complex tasks. These tasks included opening files, editing documents, and managing software workflows seamlessly. The company stated in its official announcement: “In Claude CoWork and Claude Code, you can now let Claude use your computer to handle assigned tasks, where it can point, click, and browse just like you do.”
This update builds upon a series of rapid and innovative releases launched by Anthropic this month. On March 17, the company rolled out a feature named “Dispatch” as a research preview inside the “Claude CoWork” tool, a productivity feature functioning as an autonomous agent dedicated to macOS. This feature creates a continuous and effective conversation between the user’s phone and personal computer, allowing tasks to be assigned and commands directed remotely with complete ease.
Felix Riesberg, a technical staff member at Anthropic, announced the “Dispatch” feature via the social media platform X, writing: “It’s one continuous conversation with Claude running directly on your computer. You can message it from your mobile phone and come back later to find the work completely done.”
Expanding productivity tools and security
The “CoWork” tool first appeared in January 2026 for Max subscribers before expanding its availability to Pro users. Unlike the “Claude Code” tool, which is primarily targeted at developers and programmers, “CoWork” was specifically designed for knowledge workers and office professionals.
The tool runs locally on Mac devices inside a completely isolated virtual machine. Choosing to run this feature within an isolated environment reflects Anthropic’s deep awareness of security concerns related to granting AI absolute control over user computers. This environment ensures that any unexpected behavior from the intelligent model will not affect the stability of the underlying operating system. The tool has access to local files with the ability to integrate and connect with popular work tools such as Gmail, Slack, Notion, and GitHub.
How it works and current technical limitations
Setting up the “Dispatch” feature is simple and straightforward; users only need to update the dedicated Mac app for Claude, scan a QR code to link the app to their mobile phone, and then send instructions remotely. All computing processes take place on the user’s desktop computer, and Claude requires explicit approval before executing any practical actions. Additionally, the Mac must remain awake and connected to the internet for tasks to run and finish successfully.
Despite these impressive capabilities, early practical testing conducted by the tech publication MacStories revealed that results remain mixed. While tasks such as summarizing emails and locating files succeeded reliably and quickly, tasks involving Safari browser tabs or command-line sessions frequently failed. The publication noted in its evaluation: “The feature still has a long way to go before becoming fully equivalent to operating the tool remotely,” describing the success rate as roughly 50/50.
Intense competition in the AI agent market
Anthropic’s push toward developing AI agents capable of using computers aligns with a broader trend across the technology sector. Major competing companies, including OpenAI, Google, and Meta, have developed their own agentic tools capable of operating within desktop and browser environments to meet growing user needs.
Currently, the new feature is limited to Mac devices, with no announced timeline for supporting Windows-based computers. Pricing for these services remains tied to Anthropic’s current subscription structures, with Max plans starting at $100 per month, while Pro plans cost $20 per month. It is clear that we stand on the threshold of a new era of productivity, and as companies continue to refine their algorithms, we can expect significant improvements in the success rates of these tools in the future.
Frequently Asked Questions
What is the new update launched by Anthropic for the Claude model?
Anthropic launched an update that allows the “Claude” AI model to directly control Mac computers, enabling it to click, browse, open applications, and manage files just like a human user.
How does the Dispatch remote control feature work?
The “Dispatch” feature allows users to link their mobile phones to their Mac computers by scanning a QR code, enabling them to send text commands from their phone for Claude to execute on the desktop computer.
Do these new features work on Windows devices?
Currently, these new features and capabilities are limited to Mac devices only, and Anthropic has not announced any official release date for a version supporting Windows-powered computers.
Is this feature safe for my data and personal computer?
Yes, the tool is designed to run locally inside a completely isolated virtual machine on the Mac, preventing the AI from tampering with core system files and requiring user approval before executing actions.
