Claude Cowork Breaks Free From the Desktop, Expanding to Web and Mobile
Anthropic is taking its knowledge-work AI agent off the laptop entirely, rolling Claude Cowork out to web browsers and phones so tasks can start at a desk, run unattended overnight, and finish by the time someone checks their phone the next morning.
Claude Cowork, Anthropic’s agent for handling everyday office tasks rather than just code, is moving beyond the desktop app it launched with in January.
Starting Tuesday, the tool is rolling out to web browsers at claude.ai and to the Claude mobile app for iOS and Android, according to Anthropic’s own announcement.
The expansion is available first to Max subscribers as a beta, with access to additional plans expected to follow. The change means a task that starts on a laptop no longer has to stop the moment that laptop closes.
What Actually Changes for Users
Anthropic frames the update around three shifts. Work now follows users across devices, letting them start a task on a desktop and pick up where they left off on a phone, reflecting a broader push toward seamless cross-device workflows seen in OpenClaw’s recent mobile rollout.
It also continues in the background without an active connection, so a scheduled task can work through emails, meeting transcripts, and recent news before leaving a finished briefing and a drafted follow-up email ready for review.
Crucially, Anthropic says decisions requiring human judgment are routed back to users before anything ships, meaning Claude pauses to ask rather than acting unilaterally.
Ramp customer success employee Armmand Hosseini described building a client-tracking dashboard that started on a laptop and resumed on a phone while waiting for luggage at an airport, saying the session simply “held the thread.”
The desktop app, which got more advanced plugins earlier this year, still offers the deepest experience with access to local files and the browser, while web and mobile now let people use Cowork without installing the desktop app.
This shift toward unattended execution underscores why human-in-the-loop guardrails remain essential. Without clear human oversight, even advanced AI agents risk causing software damage or executing unintended commands instead of delivering a safe, productive fix.
The Data Behind the Push
Anthropic’s own usage numbers help explain why it’s betting on Cowork beyond coding circles.
The company analyzed 1.2 million anonymized Cowork sessions from more than 600,000 organizations over the final two weeks of May.
It found that business process operations, including reconciling spreadsheets, building onboarding checklists, and combining scattered updates into a single report, were the largest category at 33.4%, concentrated across finance, HR, and administrative roles.
Content creation and copywriting followed at 16.4%, covering drafts, slide decks, and proposals typically handled by marketing and management.
Software development, by contrast, accounted for just 8.7% of total usage, a notable split for a company whose earlier agentic tools built their reputation on writing code.
A Broader Race to Own the Workday
TechCrunch framed the expansion as part of a wider trend, saying the “coding agent wars are spilling into the rest of the office,” with tools like OpenAI’s Codex expanding from software development into reports, spreadsheets, and research for non-technical users.
Anthropic has been building out this broader office footprint in other ways too, including Claude Tag, an always-on assistant embedded directly in Slack that the company introduced in June.
For both companies, the wager is the same: as AI tools mature, competitive advantage may depend less on the quality of the chatbot being used and more on whose AI agent becomes the one where the real work of running a business gets done.
