Anthropic analyzed 1.2 million Claude Cowork sessions from over 600,000 organizations, finding that about half of all usage is for mundane business processes and text creation, such as compiling status reports and building checklists. Software development is rarely done in Cowork, as developers prefer Claude Code.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Anthropic's analysis of usage of its AI assistant Cowork shows that half of the sessions are used for administrative and text tasks, while core professional work accounts for a relatively small proportion, revealing the true role of AI in daily office work.
Anthropic analyzed 1.2 million anonymous sessions of Claude Cowork from over 600,000 organizations between May 11-31, 2026, categorizing each session into 20 job categories. The results show that about half of the sessions concentrated in two major categories: business process and operations (33.4%) and content creation and copywriting (16.4%).
Anthropic describes these tasks as 'work about work'—the trivial matters that almost every office job involves but are rarely anyone's core responsibility, such as consolidating scattered status updates into reports, creating onboarding checklists, or matching spreadsheet data.
After the top two categories, the remaining categories have significantly smaller shares: software development 8.7%, DevOps and infrastructure 7.0%, research 6.4%, data analysis 5.8%, document processing 4.1%, sales operations 4.0%, personal assistant 3.8%, education 2.4%, meeting analysis 1.8%.
The low software development share validates Anthropic's hypothesis: interface influences user behavior. Developers prefer using the specialized programming tool Claude Code for coding tasks, while on Cowork they handle communication tasks surrounding programming (e.g., writing documentation, making presentations).
Anthropic illustrates how Cowork fills the gaps: lawyers can use it to format documents and archive them, freeing up time for legal analysis; hiring managers can schedule interviews and summarize feedback, focusing on candidate performance; team leads can quickly generate presentations to explain tough decisions, spending more time on the decisions themselves.
The blank screen is often the first barrier in knowledge work. Cowork reduces startup costs by transforming ideas and information into drafts. However, Anthropic cautions that these tools currently act more as assistants rather than replacements for human judgment.
Anthropic points out that its classification is based on activities, not occupations, so there are no separate marketing, finance, or HR categories; these functions are likely subsumed under business process and operations, partially explaining why that category accounts for one-third.
Additionally, the sample captures only a fixed maximum number of sessions per hour, leading to underrepresentation of peak periods. About 5% of sessions involve personal use (e.g., hobbies or chat), not purely work. All data is presented as percentages, not absolute volumes. Anthropic says it will continue updating this analysis.
This analysis is entirely based on data officially disclosed by Anthropic, involving 1.2 million sessions and 600,000 organizations, providing strong reference value. However, note that its sampling method (fixed hourly cap) and classification ambiguity (e.g., not distinguishing marketing, finance, etc.) may affect the precision of conclusions.
The success of Claude Cowork reveals that AI's role in the office is shifting from 'specialized tool' to 'general assistant.' Users need to choose the appropriate interface based on task characteristics (programming vs. office documents), but currently AI performs best on repetitive, structural office tasks.
Primary report
Primary source