Anthropic is pushing its AI agents beyond the chat window. Starting today, Claude Code and Claude Cowork can directly interact with your operating system. This update enables the models to open files, navigate browsers, and execute development tools without micromanagement.

The system prefers API connectors for services like Google Workspace or Slack. When those aren't available, the agent manipulates the UI directly. Anthropic built in permission gates before actions execute, though they still advise against processing sensitive data during this preview phase. For engineering teams, this distinction matters when handling proprietary codebases or credentials.

Access is limited to Claude Pro and Max subscribers on macOS. Since this remains a research preview, expect iterations based on engineer feedback. The update also integrates with Dispatch, Anthropic's cross-device continuity feature, letting sessions flow between phone and desktop seamlessly.

Claude Cowork, launched earlier this year, targets non-developers using the same underlying agent architecture as Claude Code. This move signals a broader shift toward agentic workflows where models don't just suggest code but implement it within your local environment. For engineering teams, the ability to automate routine file management or testing scripts directly could reduce context switching, provided security protocols keep pace with capability. As these tools mature, the line between assistant and autonomous operator will likely blur further, demanding new standards for oversight in production environments.

Source: Engadget