Computer Control: Anthropic and OpenClaw are defining the new capabilities of Agentic AI

Anthropic released the preview of a new update for its AI model Claude and added heft to making it more agentic. The AI model can now take control of your computer and work on it like a human being—moving the cursor on the screen (as humans would with a mouse), typing words, and opening and closing apps.

 Anthropic gave examples of what this meant. Claude could: “Pull together a competitive analysis using local files and connected tools, then compile it into a for­matted report; open your phone simulator, interact with the app you developed, and find UX issues; fill in a spreadsheet with data from multiple sources, format it, and save it to a shared folder.” The new capa­bilities are part of the Cowork suite and, to begin with, only for paid subscribers of its Pro and Max Plans in macOS.

This move comes amid the extraordi­nary rise of OpenClaw, an open-source agentic programme, developed by an Austrian programmer. It became so popu­lar that OpenAI hired the programmer. OpenClaw is now being used across the world to create AI agents which can do myriad tasks by taking control of the computer.

Anthropic has built in safety measures—a set of param­eters defines when the computer can be taken over and only with permission.