Anthropic analyzed 1.2 million Claude Cowork sessions across 600,000 organizations and found the tool has carved out a distinct niche: handling the administrative grunt work that clutters corporate calendars.
Half of all Cowork usage goes toward business processes and text creation tasks that Anthropic calls "the work around the work." This includes compiling status reports, building onboarding checklists, drafting meeting agendas, and assembling slide decks. These tasks sit between actual project work and strategic thinking, consuming time without producing direct business value.
The data reveals a clear division of labor within Anthropic's Claude ecosystem. Developers gravitate toward Claude Code for programming tasks, leaving Cowork to handle the administrative overhead that traditionally falls to whoever has free calendar space. This separation suggests organizations recognize each tool's strength. Claude Code handles technical depth. Cowork tackles the structured but tedious documentation and process work that bogs down knowledge workers.
The finding matters because it shows where AI generates immediate organizational value. Companies don't deploy AI to revolutionize their core workflows yet. Instead, they use it to automate the peripheral tasks that waste junior staff time and distract senior people from higher-impact work. A manager spending two hours building a deck manually becomes a problem Claude Cowork solves in minutes.
The 600,000 organization sample provides substantial evidence of real-world adoption patterns. This isn't theoretical. Organizations are actively using Cowork to offload specific classes of work, and the concentration around administrative tasks shows clear user behavior rather than aspirational marketing.
The long tail of use cases remains thin. Developers stick with specialized tools. Strategic work stays human-driven. Cowork occupies the middle ground where routine processes need execution but don't justify custom automation. That positioning makes Cowork a practical productivity tool rather than a transformative AI breakthrough, which may