Anthropic began rolling out Claude Cowork to web and mobile today, ending the tool’s confinement to a single laptop. Beta access started with Max subscribers, with broader plan availability to follow over the coming weeks. Sessions, files, and in-progress tasks now sync across claude.ai, the iOS and Android apps, and the desktop client.
Cowork is Anthropic’s delegation layer: a user hands Claude a multi-step job, spanning files, calendar, email, messaging, and connected web tools, and the agent works until the task is done. Anthropic said in a companion post that more than 90 percent of Cowork usage was not software development. Business operations and content creation, such as reconciling quarterly spend into a variance memo or turning a folder of contracts into a renewals tracker, together account for roughly half of all recorded activity.
Three mechanics change with today’s rollout:
- Portability. A task started at a desk can be checked from a phone and picked up wherever the output lands.
- Persistence. Work continues after the laptop closes, and scheduled tasks now execute with no device online at all.
- Approval gates. When Claude reaches a decision only the user can make, the question routes to their phone, and nothing ships until it is reviewed.
That third point is the actual product bet. Anthropic is separating task duration from human attention span while keeping a human in the approval loop, which is the design tension every agent vendor is racing to solve without triggering the failure mode of unsupervised agents taking costly actions unsupervised.
The company has not disclosed usage volume for the beta, retention data, or how many approval requests a typical multi-hour Cowork session generates. Those numbers would indicate whether “the decisions still come to you” holds up once task backlogs grow, or whether users start rubber-stamping approvals to keep pace with an agent that works overnight. Anthropic’s own customer quote, from a Ramp employee who tracked a client dashboard build across a laptop and a phone, is a single anecdote rather than an adoption metric.
The move puts Anthropic in more direct competition with OpenAI’s Tasks and Google’s Workspace agent features, both of which have pushed toward background and scheduled execution over the past two years without becoming defaults in most enterprise workflows. Google’s own Workspace agent push in 2024 did not report adoption figures twelve months later. Session portability across devices is the more concrete differentiator here: an agent job is only as useful as a worker’s ability to check on it between a laptop and a phone, and Anthropic is the first of the major labs to ship that combination in a public beta rather than a demo.
Anthropic also merged the chat and Cowork interfaces into one home on web and desktop, so projects and artifacts persist across both surfaces, and it extended doubled Cowork usage limits through August 5 to encourage bigger test jobs during the beta window.
For teams already running Claude on internal workflows, the immediate decision is whether to route a real overnight or multi-hour job through the beta rather than a synchronous session, since that is the only way to observe whether the approval gate scales with task complexity. Enterprises evaluating agent vendors this quarter should treat device-independent session persistence as a checklist item, not a nice-to-have, given how quickly Anthropic just made it table stakes.
Anthropic described the Claude Cowork web and mobile rollout in a company blog post published July 8, 2026.
