Anthropic has added a Cowork tab to the sidebar of the Claude mobile app on Android and iOS, letting users monitor Claude Cowork tasks running on their computer from a phone Engadget.

The mobile integration is strictly observational for now. Users can check status and approve pending actions, but they cannot initiate or automate new Cowork tasks from the phone itself — the agent still runs on, and acts against, the desktop machine it's tethered to Engadget.

That desktop tether has itself changed. Anthropic updated Cowork to support background task execution, removing the prior requirement that the host machine maintain a stable internet connection throughout a job. Previously, a dropped connection could stall or kill a long-running task; the update decouples task continuity from connectivity in a way that made the mobile monitoring feature practically useful rather than merely cosmetic.

Permissioning remains the operative safeguard. When Cowork needs authorization to proceed — writing a file, sending an email, executing a command with side effects — it pushes a notification to the user's phone for approval. Anthropic's stated line is direct: "Nothing ships until you've reviewed and approved it" Engadget. That approval gate is what turns a background-capable, internet-independent agent into something a user can leave running unattended without full delegation of judgment.

Access is staged. Max subscribers get the mobile Cowork tab first, with rollout to other Claude plans — presumably Pro and any team/enterprise tiers — planned over the coming weeks. No firm date was given for broader availability.

Anthropic also disclosed a longer-term architectural direction: Cowork and the standard Claude chatbot are to be merged into a single interface, with that unification appearing first on the web client and desktop app before, presumably, propagating to mobile. No timeline was specified for this either.

Why the connectivity change matters more than the phone tab

The headline feature here is the phone notification, but the more consequential engineering change is background execution without a persistent connection. Agentic coding and task-execution tools — Cowork, GitHub Copilot workspaces, various autonomous-agent frameworks — have generally assumed the orchestrating session stays alive: a terminal, a browser tab, a socket. Severing that assumption means Cowork tasks can now survive a laptop going to sleep, a Wi-Fi handoff, or a user closing a lid mid-job, which is the actual precondition for phone-based monitoring to be worth building at all. Push notifications for approvals are the interface layer; asynchronous, connection-tolerant execution is the plumbing that makes the interface non-trivial.

The approval-gate design is worth flagging as a deliberate trust mechanism rather than a technical footnote. Agentic AI products broadly have struggled with the tension between usefulness — which scales with autonomy — and user comfort, which scales with oversight. A phone-push approval model is a low-friction compromise: it keeps a human in the loop for consequential actions without requiring that human to sit at the originating machine. Whether Anthropic's implementation actually catches every "ships" action that matters, or whether some categories of low-risk action are auto-approved, wasn't specified in the available material and is a reasonable follow-up question for anyone piloting this in a production or enterprise context.

The planned merger of Cowork and the core Claude chat interface is the more structurally interesting signal. Anthropic has, until now, kept its agentic/task-execution product distinct from its conversational assistant — a separation that mirrors, in this author's view, the industry's broader uncertainty about whether "agent" and "chatbot" are different products or different modes of one product. Collapsing them into a single surface, starting on web and desktop, suggests Anthropic has settled on the latter framing: one Claude, with agentic capability as a mode rather than a separate app. That's consistent with how Microsoft has folded Copilot agent capabilities into its existing 365 surfaces rather than shipping a standalone agent app, and it may reduce the context-switching cost that currently makes users treat "chatting with Claude" and "delegating a Cowork task" as separate mental modes.

Staging the mobile rollout behind the Max tier — Anthropic's higher-priced subscription — follows a familiar pattern of using premium tiers as an early-access mechanism for features still being hardened at scale, rather than a permanent segmentation. Whether the phone tab's utility justifies its own subscription premium once it reaches Pro users will depend heavily on how often people actually need to approve or check a Cowork job while away from their desk, a use case that will likely prove more valuable to developers and knowledge workers running long batch jobs than to casual users.