Anthropic has introduced Cowork, a new feature that uses Claude for more than just chat conversations and programming. Cowork is based on the same fundamentals as Claude Code, but focuses on general knowledge work and a broader audience.
The feature is available as a beta version for Claude Max subscribers via the macOS desktop app.
Claude Code quickly became popular among developers and hobbyists. However, they soon began using it for more than just code, including a variety of office and organizational tasks. Anthropic developed Cowork to make this approach more accessible, without the technical barriers of Claude Code and associated protocols.
With Cowork, users can give Claude access to a specific folder on their computer and give commands in plain language. The AI can read, edit, and create files and independently perform larger tasks, such as filling out expense reports based on receipts, writing reports from separate notes, and cleaning up folders or desktops.
The difference with regular chat lies in the autonomy. In Cowork, Claude works agentically: after receiving a command, it draws up a plan itself and executes it step by step, while users can add instructions or corrections in the meantime. This approach is similar to Claude Code, but is now focused on non-technical tasks. Anthropic positions Cowork as a tool for all knowledge workers, from developers to marketers.
Some of these capabilities already existed through Claude Code or integrations with the Model Context Protocol, for example, to create files directly in other tools. Cowork brings this together in a streamlined desktop interface, with features such as parallel task execution and the ability to modify tasks during execution.
The functionality can be expanded with connectors for external information sources and with skills for creating documents, presentations, and other business files. In combination with Claude in Chrome, Cowork can also perform tasks that require web access.
Users remain in control because only explicitly authorized folders and connectors are accessible, and Claude requests confirmation for significant actions. At the same time, Anthropic warns of the risks of agent-like AI. Unclear prompts can lead to unwanted or destructive actions, such as deleting files, especially among less technical users.
Prompt injection also remains a concern. Although Anthropic has taken measures to address this, the company acknowledges that the security of autonomously acting AI is still under development.
Provisional research preview only available
Cowork is therefore currently only available as a research preview for Max subscribers, and there is no timeline for a wider rollout yet. The early release should provide insight into usage and areas for improvement. Anthropic is working on enhancements such as synchronization between devices and support for Windows.
The introduction of Cowork fits into a broader strategy in which Anthropic is expanding its focus beyond developers. Shortly before that, the company also announced Claude for Healthcare, competing with similar initiatives from OpenAI and positioning Claude more explicitly as a general work assistant for professional environments.
