Claude Cowork Becomes a Cloud-Powered, Cross-Device Office Agent
Claude Cowork is Anthropic’s delegated-work AI agent that now runs on desktop, web, and a Claude Cowork mobile app, using cloud-powered background task execution so knowledge workers can offload multi-step business tasks, monitor progress, and review results across devices without keeping a single computer awake or a desktop session open. This shift matters more than a feature update; it redefines who AI agents are built for. Anthropic first positioned Cowork as a desktop-centered coding and automation assistant, but the new web and mobile rollout is clearly designed for office staff drowning in busywork rather than engineers chained to IDEs. By letting Claude keep running in the background even when a laptop is closed and surfacing status updates on phones and browsers, Anthropic is betting that workers want an AI agent that fits the messy rhythm of meetings, commutes, and scattered workdays, not just neat desktop sessions.
Usage Data Shows Claude Cowork Is an Enterprise AI Workflow Tool, Not Just a Coding Sidekick
Anthropic’s own numbers make the strategic pivot unavoidable: in a sample of 1.2 million Cowork sessions across 600,000 organizations, software development represented only 8.7% of usage. The largest slice, 33.4%, was business process operations like reconciling spreadsheets, building onboarding checklists, and pulling scattered updates into reports. Content creation and copywriting followed at 16.4%, covering drafts, slide decks, proposals, and social posts. In plain terms, Cowork has already become an enterprise AI workflow engine: a quiet back-office agent that handles administrative processes, documents, presentations, reports, and other repeatable knowledge-work tasks where humans want to review, approve, and hand off work rather than sit through a single desktop session. One quotable conclusion from Anthropic’s own analysis sums it up: “Software development accounted for less than 9% of Claude Cowork usage before Anthropic took it mobile and web.” Given that reality, keeping the agent locked inside a desktop app made little sense.
From Desktop Prison to AI Agent Cross-Device Sync
The most important change is not where you click; it is where the work runs. Claude Cowork sessions now default to running in the cloud, instead of relying on a single machine staying on and awake. That unlocks genuine AI agent cross-device sync: you can start a task on your laptop, check its status from your phone, or pick it up later in a web browser without losing the thread. Scheduled tasks can run in the cloud even when all your devices are offline, and Claude can keep working in the background after the laptop lid closes. This is what makes Cowork feel less like a chatbot and more like a work queue. Users can let the agent collect data from email threads, messaging channels, meeting transcripts, and recent news for a deal briefing, then review the output while they are away from their desks. Cloud-powered AI agents only become useful at scale when they are untethered from a single device, and Anthropic has finally treated Cowork like a service, not a local app.
Web, Mobile, and Desktop: Dividing Access From Full Capability
Anthropic’s design choice is clear: separate access from capability. On web and mobile, users can start, steer, review, and resume Cowork tasks, use connectors, skills, plugins, scheduled tasks and Projects, and preview files Claude creates. They can approve next steps from a phone or browser, get mobile alerts when Claude needs their sign-off or has something ready for review, and return to the output when it suits them. Meanwhile, the desktop app remains the full Claude Cowork environment for local files, local connectors, browser access, and computer control. A remote session can continue after the desktop app closes, but it can no longer reach local files once that app is off. In practice, that means the desktop app is where heavy automation lives, while web and mobile become the control room. For office workers, this split is a feature, not a bug: they get fast access to the agent’s brain from anywhere, without giving a phone unrestricted control over their machines.
Safety, Pricing, and the Road Ahead for Everyday Office Agents
Anthropic is not throwing caution out the window to gain convenience. Cowork can touch calendars, email, messaging tools, browser actions, and connected services, so it relies on human approval gates even in its new cloud-powered, cross-device form. Users receive mobile alerts when Claude needs their approval or has results ready, turning human review into a central safety control rather than an afterthought. Access is also being gated commercially. The Claude Cowork mobile app and web interface are rolling out first to Max subscribers, priced at USD 100 (approx. RM460) a month, with broader plan access expected in the coming weeks. Anthropic is doubling temporary Cowork usage limits through August 5, giving early adopters room to push the agent into more of their enterprise AI workflow before tighter quotas return. The direction is unmistakable: Cowork is evolving into a background office agent that quietly runs cross-device business processes. If AI agents are going to earn their place in everyday work, this mix of cloud execution, cross-device sync, and explicit human oversight is the shape they will need to take.
