Anthropic's Own Usage Data Says Claude Cowork Is Barely a Coding Tool

Software development accounts for just 8.7% of Claude Cowork sessions, according to Anthropic’s own analysis of roughly 1.2 million sessions across more than 600,000 organizations between May 11 and May 31, 2026. The largest category by far is business process and operations at 33.4% — consolidating scattered updates into a single report, building onboarding checklists, reconciling spreadsheets — followed by content creation and copywriting at 16.4%: drafts, slide decks, posts, proposals. DevOps and infrastructure trails at 7%, research and intelligence at 6.4%, data analysis at 5.8%, document processing at 4.1%, and sales and revenue operations at just 4%. Anthropic’s own framing for this data is “the work around the work” — the tasks that span nearly every role and rarely appear in a job description, which is also the argument the company is using to steer enterprise buyers toward provisioning agent seats for finance, HR, and operations staff, not just engineering teams.

The timing lines up with Cowork’s expansion to web and mobile, announced the same week, which lets a session started at a desk get checked from a phone and finish wherever it lands. Anthropic’s example is a Monday-morning client-prep job that runs at 6 a.m. unattended, pulling emails, transcripts, and recent news into a briefing doc and leaving a follow-up email drafted but unsent for a human to approve — exactly the “business process and operations” work the usage data says is already the dominant category, not a hypothetical. Usage limits are doubled through August 5, 2026 to mark the launch.

For a consulting practice advising companies on where AI actually pays off, the number worth remembering isn’t 8.7% — it’s that the company that makes the leading coding agent is now selling its product primarily on the strength of work that has nothing to do with code. If the frontier lab’s own telemetry says the value is in operations and content, not just engineering headcount, that’s a stronger argument for where to point a first agent deployment than any vendor pitch.