Claude Cowork Escapes Desktop as Anthropic Bets on Agents

Claude Cowork leaves the desktop, and Anthropic is testing whether AI agents can become daily infrastructure

Claude Cowork is no longer just a desktop agent, it’s becoming a persistent work layer that follows users across devices and keeps running after the laptop closes.

Starting Tuesday, Anthropic is bringing Claude Cowork on mobile and web for the first time, according to The Verge. The rollout starts with Max subscribers, then expands to Claude users on other plans “in the coming weeks.”

The surface-level story is platform expansion. The deeper shift is continuity. Cowork sessions now run in the cloud by default, which means a task can begin on a desktop, continue while the user moves away, and later be checked from a phone or browser. That changes the product from a workstation tool into something closer to a standing work queue.

Anthropic’s own framing points in that direction. In its Claude product announcement, the company says Cowork is where users hand Claude a task and let it work “across your files, calendar, email, messaging app, the web, and the other tools you connect until the job is done.”

The counterpoint matters: the full Cowork experience still lives on desktop. Anthropic says the desktop app remains the place for features such as local file access. So this isn’t a clean break from the laptop. It’s a split model: desktop for deep access, mobile and web for continuity, monitoring, approvals, and lighter handoffs.

Max subscribers get Claude Cowork first as Anthropic turns access into a premium AI feature

Anthropic is using the Max plan as the first gate for Claude Cowork’s mobile and web beta. That’s not just a rollout detail. It shows how advanced agent features are being treated inside Claude: higher-friction, higher-value, and not yet ready for blanket distribution.

The company says beta access is rolling out “over the next several weeks starting with Max users, with more plans to follow.” That staged release gives Anthropic room to observe how people actually use long-running agents across devices before opening the tap wider.

XOOMAR analysis: this is a sensible way to release agentic software. Cowork isn’t a simple chat feature. It can handle multi-step tasks, work with connected tools, run scheduled jobs, and ask for human approval when needed. That creates heavier operational demands than a standard prompt-and-response session. It also raises more safety and support questions.

The result is a more meaningful subscription ladder. If Cowork becomes the version of Claude that actually performs work rather than just answering questions, access to it becomes a reason to pay up. That same pressure is visible across adjacent AI product strategy, where model choice and agent access are increasingly treated as product architecture, a theme XOOMAR has covered in Model Lock-In Cracks as Vercel AI Agents Pick Labs.

The numbers behind the Claude Cowork rollout: devices, timelines, and feature gaps

The facts are tight and specific. Claude Cowork on mobile and web begins rolling out Tuesday. Before this change, Cowork was available only through the Claude desktop app for macOS and Windows. Now it adds iOS, Android, and the web.

The platform expansion is broad, but the capabilities aren’t identical.

The most important technical change is that Cowork sessions now run in the cloud by default. That lets users continue sessions across devices and allows tasks to keep running in the background even when a laptop is closed. Anthropic also says scheduled tasks can now run even when none of a user’s devices are online.

There is still an option for local processing in the desktop app, where users can switch between cloud and local processing. That choice matters. Local processing keeps certain workflows tied to the machine, while cloud sessions make Cowork portable.

Anthropic is also extending its doubled Cowork usage limits through August 5th. That gives users more room to test bigger tasks during the rollout window, though the source material does not specify the exact usage limits.

Cloud-based Cowork sessions make Claude more useful, but they sharpen the trust problem

Cloud-default Cowork makes the product more practical. A user can start a workflow at a desk, leave for a meeting, get a phone notification, approve a step, and later pick up the output from a browser. That’s the agent promise in plain terms: work doesn’t stop just because the user changed devices.

Anthropic emphasizes that approval remains part of the loop.

“Nothing ships until you’ve reviewed and approved it.”

That sentence is doing a lot of work. It reassures users that Claude can keep moving without fully taking over. The product challenge is making that boundary visible. If Cowork handles files, messages, calendars, and connected tools, users need to know what it can access, what it changed, and when it needs permission.

Anthropic’s Help Center language is blunt about one related capability, Dispatch, which lets users message Claude from a phone and have it work on a desktop. The company warns that “instructions from your phone can trigger real actions on your computer,” including reading, moving, or deleting local files, interacting with connected services, controlling a browser, and using desktop apps.

That warning does not mean cloud-default Cowork is unsafe. It means Anthropic understands the trust problem is real. A mobile agent that can reach into work tools is useful precisely because it can do consequential things. That’s also why permission clarity, auditability, and recovery controls will matter more than flashy demos.

Readers tracking the security side of AI assistants may see the same tension in consumer protection tools like Savi AI Scam App Hunts Fake Ransom Calls Before Panic: automation is valuable only when users understand where the machine acts and where the human remains in charge.

Claude Cowork’s real test is knowledge work, not coding

Anthropic says more than 90% of Claude Cowork usage wasn’t software development. The company says most of it was everyday knowledge work, with the largest categories being business operations and content creation, which together made up roughly half of all usage.

That matters because AI agents first gained attention through coding workflows, where tasks can be tested, run, and checked in relatively structured environments. Cowork is aimed at messier office work: reconciling spend, drafting variance memos, turning contracts into renewals trackers, or building client decks from call transcripts and pipeline data.

Those examples come from Anthropic’s own product post, so they should be read as company framing, not independent usage research. Still, the claim is revealing. Anthropic is positioning Cowork less as a developer tool and more as an AI layer for the “work around the work.”

Mobile access changes expectations here. Once Cowork appears on a phone, users won’t treat it as a desktop-only assistant. They’ll expect status checks, quick corrections, task approvals, and handoffs from anywhere. The phone becomes the control surface, not necessarily the place where the hardest work happens.

Claude Cowork’s next test is whether mobile agents stay controlled as they get more useful

XOOMAR analysis: Anthropic is likely to keep desktop as the most capable Cowork environment while using mobile and web for continuity, review, approvals, and lightweight commands. That division fits the product facts. Desktop has local file access and deeper workstation context. Cloud sessions and mobile notifications make the agent persistent.

The evidence that would strengthen this thesis is straightforward: more file and app integrations on desktop, clearer session histories across devices, and stronger permission controls for cloud and local work. The evidence that would weaken it would be Anthropic bringing full local-style capability to web and mobile without keeping desktop as the main deep-work surface.

For now, Claude Cowork is moving in a careful direction. Anthropic wants the agent to keep working when the laptop closes, but it also keeps reminding users that real work requires review. That balance will decide whether Cowork becomes a useful daily work layer or another AI feature people try once and abandon when the permissions get confusing.

The Bottom Line

• Claude Cowork is shifting from a desktop tool into a cloud-based work layer that can follow users across devices.
• Anthropic is positioning advanced AI agent features as premium capabilities by giving Max subscribers first access.
• The launch highlights a broader move toward persistent AI agents that manage tasks across files, calendars, email, messaging, and the web.