Anthropic Introduces 'Claude Cowork' in New AI Push

On Monday, Anthropic unveiled an innovative feature known as 'Claude Cowork' aimed at broadening the utility of their AI agent, Claude. This feature emerges amidst heightened interest in Claude Code. The aim is to offer an intuitive platform for non-coding tasks.

Anthropic describes Cowork as a more user-friendly counterpart to Claude Code, enabling it to manage similar types of tasks. This 'research preview' phase is intended to gather data on user interaction and further refine development. Presently, Cowork can be accessed exclusively via Claude's macOS application for those subscribed to the premium Claude Max plan, which is priced between $100 and $200 monthly, depending on usage.

Functionality of Claude Cowork

Under this feature, users can grant Claude access to specific folders, empowering the AI to read, edit, or generate files. For instance, Claude can autonomously organize your downloads by renaming and sorting files, compile a spreadsheet of expenses from screenshots, or draft reports based on loose notes. Users receive ongoing updates on the AI's progress and may connect to external platforms like Asana, Notion, and PayPal, or leverage Claude's capabilities in Chrome for browser tasks.

Anthropic highlights the advantage of eliminating the repetitive task of inputting context or formatting Claude's outputs manually. Users can schedule tasks and let Claude proceed without constant oversight, mimicking more of a collaborative office environment rather than a straightforward command-response interaction.

Objective Behind the Feature

The launch of this feature aligns with Anthropic's goal, shared by competitors, to devise AI agents that deliver practical benefits to both individual users and businesses. Despite significant progress, AI agents still have strides to make before they become commonplace for everyday users.

'Skills for Claude,' another feature released in October, laid the groundwork for Cowork by allowing customization for task improvement. By using this, Claude could become adept in specific duties through folders containing instructions and scripts, ranging from adapting to brand guidelines to Excel tasks. Users could develop their tailored Skills based on particular requirements.

Safety and Considerations

Anthropic also cautions users about potential hazards when utilizing Cowork and other similar AI tools. Ambiguous directives pose risks, such as the unintended deletion of files. Moreover, prompt injection attacks—where harmful code is hidden by malevolent actors—can trigger unsafe actions by the AI. Ensuring agent security, particularly concerning Claude's real-world applications, remains a vital area in AI advancement.

Claude Max users can access this new feature by selecting the 'Cowork' option within the macOS app, while others have an opportunity to experience it actively through updated subscriptions.