Anthropic has expanded Claude Cowork, its general-knowledge-work agent, from desktop to web and mobile, rolling out in beta to Max subscribers first. A feature called Dispatch keeps one persistent thread that routes tasks to Claude Code or Cowork and lets work continue while the laptop is closed. Anthropic also released usage data showing business-process and content tasks dominate, with coding just 8.7% of sessions.TL;DR
Anthropic has brought Claude Cowork, its Claude Code-style agent for general knowledge work, to web and mobile. The tool launched as a desktop app in January and, from Tuesday, is rolling out in beta to Max subscribers first.
The pitch is continuity across devices. Users can kick off a task at their desk, get status updates on their phone, and collect the finished output later, even with the laptop closed.
Under a feature Anthropic calls Dispatch, a single persistent thread routes each request to the right engine. Development work runs in Claude Code, knowledge work runs in Cowork, and Claude messages back the result rather than every step.
The framing is deliberate, positioning Cowork less as a coding tool and more as an administrative coworker that runs in the background and pings you only when a decision is yours to make. In Anthropic’s words, the agent handles the “work around the work”.
The agent wars leave the codebase
The move fits a broader industry pivot from chatbots toward the surfaces where work actually happens. OpenAI is doing the same with Codex, expanding its coding tool into a general enterprise-work platform for non-developers.
The bet for both labs is that dominance will hinge less on the best chatbot and more on owning the space where tasks get done. Startups are crowding in too, with Viktor raising $75m to embed an AI coworker in Slack and Teams.
Anthropic has been widening Claude’s footprint across office software for months, from bringing Claude into Microsoft Word to enterprise deployments like KPMG putting Claude in front of 276,000 staff. Google is pressing the same front with its Gemini Spark agentic assistant.
Most users aren’t coding
Anthropic also released early Cowork usage data that undercuts the coding-first stereotype. It sampled 1.2 million anonymised sessions from more than 600,000 organisations across the last two weeks of May.
The biggest category, at 33.4%, was business process work like reconciling spreadsheets and building reports, common in finance, HR, and admin. Content creation and copywriting followed at 16.4%, while software development accounted for just 8.7%.
The company frames this as evidence that everyday business tasks, not coding, are where AI is landing hardest. The desktop app stays the home for deep work with local file and browser access, while web and mobile widen the door to people who never installed it.
There is a sharp edge to giving a phone remote control of a desktop agent, since a manipulated instruction or a phishing link could trigger hard-to-undo actions on a machine. Anthropic’s own guidance urges users to connect these agents only if they trust every app in the chain, a caution worth heeding as the office fills with autonomous helpers.