Anthropic launches 'Claude Cowork' to bring AI agents to desktop

Anthropic launches “Claude Cowork,” a new agentic AI feature designed for general office tasks. It builds upon the success of its developer-focused tool, Claude Code. This feature integrates into the macOS Claude desktop app and specifically targets knowledge workers, such as marketers and analysts.

Cowork is designed to be highly accessible and user-friendly. It allows users to grant the AI access to specific computer folders. They can then direct it using simple, plain-language commands. Cowork improves upon the complex Claude Code by providing a streamlined interface. Users can modify instructions mid-task without writing a single line of code.

Anthropic envisions a wide range of practical applications for Cowork. These include automating the generation of expense reports from a folder containing receipt photos.

Moreover, it can also synthesize comprehensive reports by drawing information from numerous digital notes. Additionally, it efficiently organizes and declutters desktop files.

However, the tool comes with noteworthy warnings. Anthropic advises that inadequate prompts may lead to unintended and potentially harmful consequences. These include the accidental deletion of files.

Furthermore, the AI company also acknowledges the ongoing industry-wide risk of prompt injection attacks. In these cases, external inputs can exploit the AI’s behavior.

As these risks persist, Cowork is launching strictly as a research preview available only to Claude Max subscribers, with no date set for a wider release. This announcement follows a busy week for the company, which unveiled “Claude for Healthcare” on Monday. This move is aimed at competing with OpenAI’s recent growth into health management tools.