Claude’s Game-Changing Update: Autonomous Task Execution Launches a New AI Experience

Anthropic has recently rolled out a groundbreaking feature for Claude, enabling it to autonomously execute tasks directly on users’ computers. This update is redefining the productivity boundaries of AI assistants. From automatically opening files to running development tools, Claude now works like a true “digital colleague,” moving far beyond simple conversations.

New AI Autonomous Control Capability

Claude’s new “computer operation” ability lets Claude Code and the Cowork model perform a variety of tasks directly on the user’s macOS system. This includes screen navigation (clicking, scrolling), opening files, browsing the web, and even running development tools—all without any special integration setup.

How practical is it? Imagine assigning a task to Claude via the Dispatch service on your phone, and it automatically completes it on your computer. This seamless cross-device collaboration delivers a massive leap in productivity. Currently, the feature is available as a “research preview” to Claude Pro and Max subscribers.

Enhanced Developer Experience with Auto Mode

For developers, the new “auto mode” in this update is a game-changer. In auto mode, Claude Code can not only write code but also execute a variety of Shell commands—such as creating, moving, and deleting file directories, and even pushing code updates to GitHub.

More importantly, the built-in AI safety classifier automatically reviews each tool call before execution, preventing destructive actions like mass file deletion or sensitive data leakage. Operations deemed safe are executed automatically, while risky commands are blocked and prompt Claude to reconsider. If violations persist, user confirmation is required—significantly reducing the need for manual intervention.

Simplified Interface and Safety Assurance

A standout feature of this update is the dramatically lowered barrier to entry. Users don’t need to learn complex APIs or perform deep configuration—Claude’s autonomous operation is nearly plug-and-play. Anthropic emphasizes that the system requires explicit user approval before launching any new application, and users can abort execution at any time.

It’s worth noting that while newer models like Claude Sonnet 4.5 deliver top-tier performance on coding benchmarks, Anthropic acknowledges that the computer operation feature is still in its early stages and advises users to avoid handling sensitive information. Currently, auto mode is available to Team plan users, with plans to gradually extend it to Enterprise and API users.