The move comes at a time when AI agents are quickly replacing traditional chatbots. Instead of just answering questions, these systems are now expected to act. In a statement, Anthropic described the feature simply: “When Claude doesn’t have a connector or tool for what you need, it navigates your screen directly—clicking, typing, and opening apps just like you would.”
Claude moves from chat to action
Claude’s new capability works in layers. First, it tries to use direct integrations like Gmail or Slack. If those are not available, it shifts to browser control. As a last step, it interacts with the screen itself.
That means the AI can now open files stored on your system, fill spreadsheets and save them, use developer tools or simulators, and also navigate internal dashboards without APIs.
Anthropic says the system “uses the most precise tool first,” making it faster when proper integrations exist.
Many developers have been experimenting with similar ideas through open-source tools like OpenClaw. Now, big AI firms are bringing that concept into mainstream products.
Permissions, risks, and real concerns
Giving an AI access to your computer sounds useful, but it comes with risks. Anthropic itself admits the system is not perfect.
The company notes that Claude can see everything on your screen during operation. That includes sensitive files, emails, and personal data. “This means Claude can see any information visible on your screen,” the company warned.
To reduce risks, the system asks for permission before accessing apps. Some apps are blocked by default. Users can also stop the AI at any time.
Still, experts remain cautious. Agentic systems can act fast, sometimes without clear user intent. If compromised, they could misuse data or perform unwanted actions.
Rollout plans
Right now, the feature is only available on macOS through Claude’s desktop app. Windows support is expected later. Anthropic says complex workflows may fail and might need a second try.
There’s also a simple requirement that feels oddly human: your computer must stay awake, as the AI cannot work if the screen sleeps.
And if this works at scale, the way we use computers might quietly change.