Anthropic has published usage data drawn from 1.2 million anonymised Claude Cowork sessions across more than 600,000 organisations, and the breakdown cuts against the assumption that AI coding tools are mainly an engineering department’s concern. Business process and operations work — pulling scattered updates into a report, building onboarding checklists, reconciling spreadsheets — is the single largest category at 33.4% of sessions. Content creation and copywriting comes second at 16.4%. Software development accounts for just 8.7%. The same pattern shows up at Epic Systems, the healthcare software company, where Anthropic reports that over half of internal Claude Code usage now comes from non-developer roles — including a pharmacist who built a fully interactive prototype without writing a line of code themselves.
The read-across for anyone commissioning software isn’t that coding tools have become unimportant — Claude Code and its peers are still the fastest-growing category in the AI tooling market. It’s that the buyer for AI-assisted development capability inside a business is no longer just the CTO evaluating a coding assistant. It’s operations leads, clinical staff, finance teams and product managers who now expect to be able to describe what they want and get something functional back, and who are increasingly the ones deciding whether a business’s internal tools get built properly or get left as a patchwork of AI-generated prototypes nobody owns.
So what
That shift raises the stakes on what “AI-assisted” should mean for a commissioned build: not just faster developer output, but software and internal tools designed so non-technical teams can extend or request changes safely, without every prototype becoming unmaintained shadow IT. If your organisation already has non-developers building things with AI tools and nobody has thought about where that meets production-grade software, that’s worth a conversation before it becomes a support burden. See our AI-assisted development work, or get in touch to talk through turning ad hoc AI prototypes into something your team can actually rely on.