Anthropic has released Cowork, a new feature that extends Claude’s capabilities beyond coding to general knowledge work. The tool is available as a research preview for Claude Max subscribers on macOS.
Cowork allows users to grant Claude access to specific folders on their computer. The AI assistant can then read, edit, and create files within those directories. According to Anthropic, Claude can reorganize downloads, create spreadsheets from screenshots, or draft reports from scattered notes.
The feature operates with greater autonomy than standard chat interactions. Users assign tasks, and Claude develops a plan and executes it while providing progress updates. Anthropic states that Cowork is built on the same foundation as Claude Code, which was initially designed for developers.
Cowork can create sub-agents to handle complex tasks in parallel. Each sub-agent starts with fresh context, which prevents the system from hitting context limits during extended work. Users can queue multiple tasks and check back later for completed results.
The company emphasizes user control. Claude requires explicit permission to access folders and asks before taking significant actions. However, Anthropic warns that Claude can perform destructive actions like deleting files if instructed, and users should provide clear guidance. The company also notes risks from prompt injections, where malicious content could alter Claude’s behavior.
Current limitations include macOS-only availability, no cross-device synchronization, and missing features like Projects and Memory. Anthropic plans improvements based on user feedback from this preview phase.
Sources: Claude Blog, Claude Resources
