Anthropic is rolling out a major upgrade to its AI assistant: Claude can now actively use your computer to complete tasks, not just suggest steps. Through the Cowork and Code experiences in the macOS desktop app, Pro and Max subscribers can ask Claude to open apps, browse the web, navigate files, and carry out multi-step workflows with on-screen control—no extra setup required.
What Claude Can Do On Your Desktop With Computer Use
The new “computer use” capability lets Claude handle common digital chores end-to-end. Think filling out web forms, pulling numbers from PDFs into a spreadsheet, organizing folders into project-based categories, or grabbing reference material from multiple tabs and assembling it into a draft document. You describe the outcome; Claude plans the steps and executes them on your screen.
Anthropic routes actions through connected services first—such as Slack or Google Calendar—if a connector fits the job. When no connector exists, Claude switches to direct control, using your mouse, keyboard, and browser to complete the task. You see it scroll, click, and type in real time, and it asks for explicit permission before accessing any file, folder, or app you haven’t approved.
It’s still early. Anthropic cautions that the agent can make mistakes, and it recommends starting with non-sensitive workflows. You can stop an automation at any time, and every session is bounded by the permissions you grant. In practice, that means you can safely test Claude on routine admin work—like renaming files or collating meeting notes—before letting it touch important data.
How Dispatch Extends Control From Your Phone
A standout addition is Dispatch, a mobile-first control panel that lets you instruct Claude to operate your desktop remotely. As long as your Mac is awake and online, you can be on your phone and ask Claude to check Slack, skim emails for action items, pull figures from a local spreadsheet, or even start a presentation draft using files on your machine.
Dispatch maintains a single persistent thread so the agent remembers prior tasks and your preferences. That memory is user-editable—you can view, adjust, or delete it at any time. For recurring work, you can also ask Claude to schedule tasks or compile daily and weekly summaries drawn from your files and connected apps.
Safety and Permissions by Design for Secure Computer Use
Anthropic frames the feature around consent and transparency. Before Claude touches new resources, it requests approval. While it works, you can watch each step as it clicks and types, creating a visible audit trail you can interrupt instantly. Access is scoped to what you allow, and nothing outside those bounds is reachable.
For organizations, this model maps cleanly to existing device-management and data-handling policies: start with low-risk apps, restrict protected folders, and log agent activity. Governance guidance from frameworks such as the NIST AI Risk Management Framework underscores these practices: minimize access, verify intent, and preserve oversight throughout automated sequences.
How It Compares to Rival Agents in Desktop Automation
Agentic “computer use” is emerging fast across the industry. OpenAI has previewed similar screen-control capabilities for ChatGPT, and Meta-backed teams are exploring PC-controlling assistants following recent acquisitions in the space. A viral community project, OpenClaw, highlighted just how compelling mouse-and-keyboard automation can be for everyday tasks.
Anthropic’s approach stands out for two reasons: it’s shipping in a consumer-ready desktop app with minimal configuration, and Dispatch adds a practical mobile remote for on-the-go control. Compared with classic RPA tools or enterprise automation suites, Claude emphasizes natural-language directives and safe, incremental permissions rather than heavy upfront scripting.
Where It Helps Today for Knowledge and Technical Work
Early adopters are likely to be knowledge workers drowning in repetitive, cross-app chores. Picture an operations lead asking Claude to gather vendor invoices from email, reconcile totals in a spreadsheet, and file everything into dated folders. Or a recruiter having Claude triage candidate emails, update a tracker, and draft follow-ups based on resume highlights.
On the technical side, the updated Claude Code experience can assist with project housekeeping: opening a repo, reviewing documentation in the browser, filing issues, and assembling a changelog from recent commits. None of this replaces expert review, but it compresses the busywork that usually bookends real problem-solving.
The bottom line: Claude is moving beyond chat into hands-on assistance. With cautious scoping and clear permissions, its new computer-use and Dispatch features can offload a surprising share of digital drudgery—while keeping you firmly in the driver’s seat.
