Skip to main content
TechnologyJul 8, 2026· 3 min read

Claude Cowork Arrives on iOS, Android, and Web: Anthropic's Agent Follows You Everywhere

Starting Tuesday, Claude Cowork, Anthropic's agent for knowledge work, is now available on web and mobile devices, in beta for subscribers of the Max plan. Until now, it could only be accessed from the desktop app for macOS and Windows; now the agent can be started and monitored from iPhone, iPad, and Android, in addition to claude.ai, with other plans to follow 'in the coming weeks'.
We had already hinted at the focus on smartphones when the first signs of Cowork's mobile launch emerged. The product logic remains unchanged: you give Claude a task, and the agent works on files, calendar, email, messaging apps, and the web until the job is done. What changes is where all this happens.
Sessions now run in the cloud by default, so they can be resumed from another device or continued while the laptop is closed. Scheduled tasks are executed even without any device online: you set up a meeting prep for six in the morning, and Claude, working on email threads, transcripts, and recent news, prepares the summary document, leaving the follow-up email drafted but unsent. When the agent arrives at a decision that is solely up to the user, the question reaches the phone with a notification. "Nothing is delivered until you have reviewed and approved it," summarizes the company.
Despite this, the desktop remains the place for the complete experience, the only one with direct access to local files and the browser; however, those who hadn't installed the app can now use Cowork too. In parallel, Anthropic has unified chat and Cowork into a single screen on web and desktop, with a single sidebar, a single search, and a single space for Projects and Artifact, as well as extending the usage limits that have doubled until August 5 to accompany the launch.
Beyond the terminal
With the announcement, Anthropic also released the first usage data of Cowork, which paints a picture that challenges the stereotype of the agent being solely for developers. From a sample of 1.2 million anonymized sessions collected between May 11 and 31 from over 600,000 organizations, the most common category is business processes and operations at 33.4%: consolidating scattered updates into a single report, building lists for onboarding new hires, reconciling spreadsheets. This is followed by content creation and copywriting at 16.4%, while software development stops at 8.7%.
The two leading categories alone account for about half of the usage, and this is connective material: what Anthropic calls "the work around work", cross-role tasks that rarely constitute the core. The company interprets the data as a signal that people use Cowork to assemble and structure information upon which they then apply their expertise, leaving the repetitive part to the agent.
The race of agents beyond code
This move fits into a broader industry shift, away from chatbots towards the surfaces where the actual work happens. OpenAI follows a similar trajectory with Codex, which started as a development tool and is increasingly used by non-programmers for reports, presentations, and data analysis. The shared bet is that competitive advantage will depend less on who has the best chatbot and more on who occupies the space where tasks are completed.
In the same direction, Anthropic had already introduced an always-on Claude inside Slack, which we discussed recently. The routing of requests between Claude Code and Cowork goes through a feature called Dispatch, a persistent thread that directs every request to the right engine and returns the result instead of every single step.
However, it should be noted that giving a phone remote control over an agent operating on a machine opens up space for real risks: a manipulated instruction or a phishing link could trigger actions that are hard to reverse. Anthropic itself advises connecting these agents only when one trusts every app in the chain.
It is worth keeping in mind what does not change: the mobile app remains primarily a way to start and monitor tasks, rather than a tool to run the agent directly on the smartphone. Heavy lifting continues to live in the cloud or on the desktop, where Claude sees local files and the browser.