Claude Cowork comes to mobile and web
Anthropic's office agent now runs cross-device and continues tasks after the device goes offline, rolling out to Max subscribers first through August 5.
Anthropic launched Cowork on web and mobile on July 7, moving the agent product off its original desktop-only footprint. Tasks now run in the background after the initiating device goes offline, and Claude pushes mobile notifications asking for approval before taking irreversible actions.
Beta access goes to Max subscribers first. Anthropic extended doubled usage limits through August 5 to mark the launch.
What changed
The desktop application remains the option for work requiring local file access or browser control. Web and mobile add asynchronous reach: a task started on a work laptop can be reviewed from a phone after the session closes. Chat and Cowork now share a single interface, with projects and artifacts accessible across both.
Anthropic published anonymized data from 1.2 million Cowork sessions across 600,000 organizations. Business operations (reports, checklists, spend reconciliation, contract review) accounted for 33.4% of usage. Content creation (drafts, presentations, proposals) was 16.4%. Software development was 8.7%. The product landed primarily with knowledge workers, not developers, despite its origins as a coding-adjacent agent.
What this means for operators
An agent that runs after the device goes offline is closer to a scheduled task or a worker process than a chat session. The approval-before-action loop keeps human oversight in the chain, but it is a mobile notification, not a full review. Check how tight that gate is before routing anything you would not want reversed.
The Claude Apps Gateway from last month handles policy and spend controls for teams running Claude on cloud infrastructure. The manual mode default in Claude Code 2.1.200 is a parallel move: Anthropic is building background execution with gated approvals, not full autonomy.