Anthropic has launched what it calls the most ambitious consumer AI agent to date, giving its Claude chatbot the ability to directly control a user’s Mac??licking buttons, opening applications, typing into fields, and navigating software on the user’s behalf. The features, available as a research preview for paying subscribers, transform Claude from a conversational assistant into a remote digital operator.
The new capabilities arrive inside both Claude Cowork, the company’s agentic productivity tool, and Claude Code, its developer-focused command-line agent. Anthropic is also extending Dispatch?? feature introduced last week that lets users assign Claude tasks from a mobile phone??nto Claude Code for the first time.
How Computer Use Works
The computer use feature operates through a layered priority system that reveals how Anthropic is approaching reliability versus reach. When a user assigns Claude a task, it first checks whether a direct connector exists??ntegrations with Gmail, Google Drive, Slack, or Google Calendar. These connectors provide the fastest and most reliable path to task completion.
If no connector is available, Claude falls back to navigating Chrome via Anthropic’s browser extension. Only as a last resort does Claude interact directly with the user’s screen??licking, typing, scrolling, and opening applications the way a human operator would.
This hierarchy matters significantly: “pulling messages through your Slack connection takes seconds, but navigating Slack through your screen takes much longer and is more error-prone,” according to Anthropic’s documentation. Screen-level interaction is the most flexible mode but also the slowest and most fragile.
When Claude does interact with the screen, it takes screenshots of the user’s desktop to understand what it’s looking at. This means Claude can see anything visible on the screen, including personal data, sensitive documents, or private information. Anthropic has trained Claude to avoid engaging in stock trading, inputting sensitive data, or gathering facial images, but the company acknowledges these “guardrails are part of how Claude is trained and instructed, but they aren’t absolute.”
Dispatch: Your iPhone as a Remote Control
The strategic centerpiece may be Dispatch, which creates a persistent conversation between Claude on your phone and Claude on your desktop. Users pair their mobile device with their Mac by scanning a QR code, then text instructions to Claude from anywhere. Claude executes those instructions on the desktop??hich must remain awake and running the Claude app??nd sends back results.
Anthropic envisions use cases ranging from having Claude check email every morning and pull weekly metrics into a report template, to organizing cluttered folders or compiling competitive analyses from local files. Scheduled tasks allow users to set a cadence once and let Claude handle the rest without further prompting.
The combination of Dispatch and computer use represents a paradigm shift, according to Anthropic: “Claude can use your computer on your behalf while you’re away.” Users have already begun experimenting with combining these features with scheduled tasks, creating what one developer described as “a background worker that can interact with any app on a cron job??hat’s not an AI assistant anymore, that’s infrastructure.”
Competitive Landscape: OpenClaw and the Agent Wars
Anthropic’s timing is clearly driven by competitive pressure. The company is shipping computer use capabilities into a market reshaped by the viral rise of OpenClaw, the open-source framework that enables AI models to autonomously control computers. OpenClaw exploded earlier this year and proved users wanted AI agents capable of taking real actions on their computers.
The framework spawned an ecosystem of derivative tools??hat the community calls “claws”??hat transformed autonomous computer control from a research curiosity into a product category. NVIDIA entered the fray with NemoClaw, its own framework designed to simplify OpenClaw deployment with added security controls. Smaller startups like Coasty are also competing in the space, marketing “full browser, desktop, and terminal automation with a native experience.”
Reuters has reported that OpenAI is actively courting private equity firms amid an “enterprise turf war with Anthropic.” Both companies are competing for enterprise customers, and the ability to offer agents that can operate within a company’s existing software stack is increasingly the differentiator.
Security Considerations and Early Testing
Computer use runs outside the virtual machine that Cowork normally uses for file operations and commands. This means Claude is interacting with the user’s actual desktop and applications??ot an isolated sandbox. A misclick, a misunderstood instruction, or a prompt injection attack could have real consequences on a user’s live system.
Anthropic has built several layers of defense: Claude requests permission before accessing each application, sensitive apps like investment platforms are blocked by default, users can maintain a blocklist, the system scans for prompt injection, and users can stop Claude at any point. Early hands-on testing suggests the feature works well for information retrieval but achieves approximately 50% success rates for more complex multi-step workflows.
The company explicitly warns users not to manage financial accounts, handle legal documents, process medical information, or interact with apps containing other people’s personal information using computer use. Enterprise customers should note that audit logs and compliance APIs do not currently capture Cowork activity?? gap that could be problematic for regulated industries.
As one observer noted: “when the agent IS the user (same mouse, keyboard, screen), traditional forensic markers won’t distinguish human vs AI actions.” The question of how to maintain appropriate audit trails for AI agents that operate identically to human users remains an open challenge that the industry will need to address as these capabilities become more widespread.