Anthropic expands Claude with new Cowork feature
Anthropic has unveiled Cowork, a new feature that extends the capabilities of Claude beyond chat interactions and programming tasks.
While built on the same principles as Claude Code, Cowork targets general knowledge work and caters to a wider range of users, News.Az reports, citing foreign media.
The feature is currently available in beta for Claude Max subscribers through the macOS desktop app.
Claude Code quickly gained popularity among developers and hobbyists, who began leveraging it for more than coding, using it for a variety of office and organizational tasks. To make this functionality more accessible without the technical hurdles associated with Claude Code and its protocols, Anthropic developed Cowork.
With Cowork, users can grant Claude access to a specific folder on their computer and issue commands in natural language. The AI can read, edit, and create files while independently handling larger tasks, such as compiling expense reports from receipts, drafting reports from multiple notes, or organizing folders and desktops.
What sets Cowork apart from standard chat interactions is its autonomy. In this mode, Claude operates agentically: it creates a plan for a command and executes it step by step, while users can intervene with additional instructions or corrections. This approach mirrors Claude Code’s functionality but is now optimized for non-technical tasks. Anthropic presents Cowork as a versatile tool for all knowledge workers, from developers to marketers.
Some of these abilities were previously accessible through Claude Code or via integrations with the Model Context Protocol, such as creating files directly in other applications. Cowork consolidates these capabilities into a streamlined desktop interface, featuring parallel task execution and the ability to adjust tasks during operation.
The system’s functionality can be further extended with connectors to external information sources and skills for producing documents, presentations, and other business materials. When used alongside Claude in Chrome, Cowork can even handle tasks that require internet access.
Users maintain control since only explicitly authorized folders and connectors are accessible, and Claude seeks confirmation for major actions. At the same time, Anthropic cautions about the risks associated with agent-like AI. Ambiguous prompts could result in unintended or harmful actions, such as file deletion, particularly among less experienced users.
Prompt injection also remains a concern. While Anthropic has implemented safeguards, the company acknowledges that securing AI that acts autonomously is an ongoing challenge.
Similar news
