For years, AI has been a world-class talker. Today, it officially became a world-class doer.
Anthropic has announced a massive leap in its “AI agent” strategy, rolling out a groundbreaking Computer Use capability for Claude. While previous AI updates focused on making the model “smarter” at chatting, this update gives Claude the metaphorical “keys to the kingdom”: the ability to move the cursor, click buttons, type text, and navigate a desktop interface just like a human.
Combined with the new Dispatch feature, Claude is transitioning from a tab in your browser to an autonomous coworker that can finish your to-list while you’re away from your desk.
Moving Beyond Chat: How Computer Use Works
The core of this "agentic" push is a shift in how AI interacts with software. Traditionally, AI requires an API (a back-door connection) to talk to other apps. If an app doesn't have an API, the AI is stuck.
Claude’s new capability bypasses this limitation by using the screen itself.
- Visual Reasoning: Claude takes frequent screenshots of your desktop to "see" where buttons and text fields are.
- Input Simulation: It translates your high-level goals into mouse movements, clicks, and keystrokes.
- Hybrid Execution: Claude first tries to use direct "connectors" (like Slack or Google Calendar). If no connector exists, it "looks" at the screen and manually operates the app to get the job done.
The "Late for a Meeting" Scenario: > Imagine you're stuck in traffic. You can message Claude from your phone: "I'm running late. Find the pitch deck on my desktop, export it as a PDF, and attach it to the 2 PM calendar invite." Claude will wake up your Mac, find the file, open the PDF exporter, navigate to your calendar, and send the invite—all while you're still behind the wheel.
Dispatch: The Remote Control for Your Workflow
The "Agent Push" isn't just about what Claude can do on a PC; it's about how you trigger it. Anthropic’s new Dispatch feature allows users to assign these complex workflows via the Claude mobile app.
- Asynchronous Productivity: You assign the task on your phone, and Claude executes it on your desktop. You get a notification once the "work" is finished.
- Scheduled Autonomy: Users can now set recurring "Agentic" tasks, such as "Every Friday at 4 PM, pull the latest sales metrics from the CRM, format them in Excel, and Slack them to the team."
The "OpenClaw" Factor: Why Now?
Industry analysts note that this move is a direct response to the viral success of OpenClaw, an open-source framework that has dominated the AI agent space in early 2026. While OpenClaw offered similar power, it often required significant technical setup. Anthropic is betting that a polished, integrated "one-click" solution within the Claude ecosystem will win over the mainstream enterprise market.
Safety and the "Human-in-the-Loop"
Giving an AI control over your cursor naturally raises eyebrows. Anthropic has implemented several guardrails to prevent "rogue" actions:
- Permission-First: Claude will ask for explicit permission before accessing a new application for the first time.
- Visual Transparency: Users can see a live plan of what Claude intends to do before it starts clicking.
- Local History: To protect privacy, action logs and screenshots are stored locally rather than synced to Anthropic’s central servers.
Current Limitations
This is a Research Preview, and Anthropic is being candid about the "rough edges":
- Mac First: Currently, full screen control is limited to macOS users. Windows support is reportedly "coming soon."
- The "Slow" Factor: Navigating via screenshots is naturally slower than direct API integrations.
- Precision: Complex tasks involving "drag and drop" or high-speed scrolling can still trip the model up.
Availability
The Computer Use and Dispatch features are rolling out today to Claude Pro ($20/mo) and Claude Max subscribers. To get started, users need to update their Claude desktop app and enable the "Computer Use" toggle in the settings menu.