Update, April 3, 2026: The innovative feature allowing Claude to control user desktops is now accessible to Pro and Max users within the Windows application for both Claude Code and Claude Cowork.
Original article from March 24, 2026: Anthropic has unveiled a groundbreaking capability for its AI assistant, Claude, enabling it to directly manage tasks on a user’s computer—tasks that typically require human intervention. This development signifies a notable shift in how AI can integrate into daily workflows.
According to Anthropic, Claude initially attempts to utilize existing integrations such as Slack and calendars. It only resorts to taking direct control of the desktop when no other interface is available, positioning this feature as a supplementary option rather than a primary one. Currently, this functionality is available as a research preview in Claude Cowork and Claude Code, specifically for macOS users.
This advancement marks a significant step for Anthropic in developing an AI agent that operates within a user’s digital ecosystem, moving beyond mere text generation. Alongside this desktop control feature, Anthropic has also introduced “Dispatch,” which empowers users to remotely manage their computers from any location.
Vercept AI acquisition laid the groundwork for desktop control
The new desktop control feature is partially derived from Anthropic’s recent acquisition of Vercept AI, a startup specializing in AI-driven computer management. Co-founder Kiana Ehsani noted that her team successfully launched their first product less than a month after joining forces with Anthropic. Initially, there were concerns that the acquisition might hinder progress; however, Ehsani asserts that the opposite has occurred. She attributes this rapid advancement to Anthropic’s dynamic culture, stating, “Everyone moves fast, everyone is incredibly smart, humble and supportive, and it’s really easy to get things done.” In her perspective, the true competitive edge of Anthropic lies in its talented workforce.
Full computer use by AI still has a lot to prove
Despite the promising potential of AI managing full computer operations, significant questions remain regarding its reliability and security for practical applications beyond the research preview phase. Granting an AI system control over an entire desktop raises critical issues surrounding data privacy, error rates, and overall controllability—challenges that Anthropic will need to address before considering a wider rollout.
Previous attempts at similar features, particularly those limited to browser functionalities, have struggled to establish reliability, as evidenced by OpenAI’s experiences with its ChatGPT operator. Anthropic’s ambitious approach to full desktop control presents a broader scope of possibilities, but it also introduces a more extensive attack surface that must be navigated carefully.
