Anthropic has launched its Claude Cowork agent on mobile and the web, opening autonomous task execution to a far wider audience beyond developers. The expansion arrives alongside internal usage data revealing that knowledge work dominates the platform, not coding.

What You Need to Know

The rollout begins in beta with Max subscribers before expanding to additional plans. Tasks now sync across devices and run autonomously even when the user closes the app. Usage data from 1.2 million sessions shows business process work accounts for 33% of activity while software development makes up just under 9%. Anthropic frames this as capturing what it calls the work around the work facing knowledge professionals rather than replacing engineers.

Anthropic analyzed 1.2 million anonymized Claude Cowork sessions collected between May 11 and May 31 from more than 600,000 organizations. Business process and operations led at 33.4% followed by content creation and copywriting at 16.4%. Together those two categories account for roughly half of all sessions. Software development registered only 8.7% trailing behind categories such as DevOps and infrastructure at 7% and research and intelligence at 6.4%.

  • Business process and operations: Users pull scattered updates into reports and build onboarding checklists. These tasks made up the largest single category.
  • Content creation: Drafting slide decks proposals and posts accounted for less than half of business process volume.
  • Software development: Despite media attention coding represented a small fraction of actual usage. Claude Code remains preferred for developers.

Anthropic describes these dominant use cases as connective work that moves projects forward without being anyone's core job description. The company argues the most valuable current application is handling everything professionals do around their actual expertise rather than replacing it outright.

Mobile Access Reframes the Agent Experience

Cowork sessions now sync across devices allowing a user to start work on a laptop check progress on a phone and retrieve finished output from any device. More notably tasks can run autonomously in the background even when no device is online. Anthropic offers the example of scheduling Monday morning client preparation for 6 a.m. with Claude working through emails transcripts and recent news to build a briefing document leaving drafts unsent until the user reviews them over coffee.

When Claude encounters a decision requiring human judgment it surfaces the question to the user's phone. Nothing ships until approval according to Anthropic. Desktop remains the richest surface with local file access while the web version opens Cowork to enterprise users whose IT departments control software installation. Chat and Cowork now share a single home screen on both web and desktop.

The company is extending doubled usage limits through August 5 to encourage adoption.

Why This Matters

Anthropic is executing a two-track strategy separating developer tools like Claude Code from the broader workplace agent. The usage data confirms the larger market lies not in writing code but in the administrative overhead that consumes much of a professional's day. By enabling background execution on mobile Anthropic positions Cowork alongside capabilities once reserved for virtual assistants. Competitors including OpenAI and Google offer similar autonomous features but the company now has real usage numbers to validate the direction. For enterprise customers the implication is clear: the return on investment for AI agents may come from streamlining cross-functional coordination not accelerating engineering output.

Tactical Implications for Enterprises

The data forces a reassessment of how AI agents get deployed inside organizations. If knowledge work drives more value than code generation procurement teams may need to prioritize tools designed for information synthesis rather than software creation. The mobile and web rollout lowers the barrier for non-technical staff creating a more level playing field across roles. Users who never open a terminal can now delegate multi step tasks to an agent that runs independently and reports back on demand.