Claude Cowork Expands Beyond Coding: What the Usage Data Actually Shows

Anthropic's Claude Cowork became available on web and mobile for Max subscribers on July 7, 2026, extending an agent that was limited to desktop computers when it launched in January TechCrunch.

The update solves a practical problem. You can now start a task on your laptop, check how it's progressing from your phone, and download the results later — without leaving your original machine running the whole time. That's because Cowork now lives across multiple devices and runs in the cloud, rather than depending on a single computer to stay powered on and connected TechCrunch.

This release arrives shortly after Anthropic launched Claude Tag in June 2026, an always-on Claude instance embedded inside Slack that works more like a persistent team member than a tool you summon when needed TechCrunch. Together, these two products signal where Anthropic wants Cowork to go: an AI assistant that handles routine work continuously across devices, available to any employee, not just software developers.

What people are actually using it for

When Anthropic published usage data from 1.2 million anonymized Cowork sessions across more than 600,000 organizations in late May 2026, one pattern stood out TechCrunch.

Business process work led by a wide margin. It accounted for 33.4% of sessions — tasks like pulling status updates from scattered emails into a single report, building onboarding checklists, or matching expense receipts to spreadsheet entries. These are the kinds of recurring, tedious jobs that don't justify hiring someone full-time but eat up hours anyway TechCrunch.

Content creation and copywriting came second at 16.4%, covering drafts, presentation decks, social posts, and proposals. Software development — the category that coding-agent products have been marketed around since 2024 — accounted for only 8.7% of sessions TechCrunch.

That split is striking. Cowork's underlying technology comes directly from the same engineering tools that Anthropic, OpenAI, and other labs have spent two years building. Those original tools were designed to handle software development work — reading code files, running tests, fixing problems automatically, iterating until a task succeeds. Yet the people actually using Cowork are operations staff, marketing teams, and administrative workers who have never written code.

The broader picture is this: the term "coding agent" may have narrowed how people think about what these systems can do. An agent built to refactor multiple files of code and debug failures turns out to be well-suited to gathering information from different sources, applying judgment, and producing organized output — the same underlying pattern, regardless of whether the input is code or email or spreadsheet rows. The work only looks different; the mechanical problem is the same.

There is one important caveat worth keeping in mind. Anthropic's data comes from people who labeled their own work on the platform during a two-week window in May 2026 — right after Cowork launched as a desktop-only tool and months before this week's mobile and web expansion. Early adopters who installed a new desktop app may use it differently than mainstream employees accessing it from a browser. Once the tool reaches a broader audience, usage patterns could shift.

Anthropic's strategy with Cowork and Tag also hints at how the company sees AI assistants fitting into work. Tag is passive — it lives inside Slack and participates in conversations when relevant. Cowork is active — you give it explicit jobs and collect the results later. Other competitors like OpenAI, Google, and a growing number of startups are moving in similar directions, trying to extend these kinds of capabilities beyond engineering teams. The usage breakdown Anthropic just released gives all of them a concrete benchmark: if a product built for coding work shows its heaviest use in operations and administrative tasks, that changes how you think about where the real business value lies.

For IT leaders and managers deciding where to invest in AI agents inside their organization, the practical implication is straightforward. If the largest share of work is flowing toward business operations and content creation rather than software development, then agent tools probably belong in the budget for administrative efficiency and operations, not just in the engineering department. That's a conversation many organizations have not yet had internally.