Claude Cowork is being pulled into the parts of office life that often sit between job descriptions: reports, checklists, presentations, spreadsheet reconciliation, drafts, and summaries. According to Anthropic, that routine layer of work has become the tool’s dominant use case.
The company’s analysis is based on a sample of 1.2 million anonymized sessions from May 11 to 31, 2026, across more than 600,000 organizations. An automated system placed each session into one of 20 work categories, giving Anthropic a broad view of how people are actually using its chat-based agent.
Office coordination leads the usage
The clearest finding is that Claude Cowork is not mainly being used for glamorous or highly specialized work. It is being used for the administrative business processes and text-based tasks that keep organizations moving.
Anthropic describes about half of usage as going toward "the work around the work." That phrase covers tasks that are necessary, recurring, and often time-consuming, even when they are not the central responsibility of any one role.
The largest category in the analysis is "Business process and operations" at 33.4 percent. Anthropic says this includes work such as turning scattered status updates into one report, creating onboarding checklists, and reconciling spreadsheets.
The second-largest category is "Content creation and copywriting" at 16.4 percent. That includes drafts, slide decks, posts, and proposals. Together, those two categories account for roughly half of Claude Cowork usage in the sample.
That pattern suggests a practical reason people reach for an AI agent: many office tasks begin with unstructured information and end with a structured artifact. A report, checklist, presentation, or proposal may not be the final decision, but it often makes the decision easier to share, review, or act on.
Why these tasks suit an AI agent
Anthropic calls these activities "connective in nature." They sit between people, teams, documents, and decisions. The work often involves gathering information, reshaping it, and presenting it in a form someone else can use.
A spreadsheet can bring scattered data together. A slide deck can translate a decision for people with different levels of context. An onboarding checklist can package institutional knowledge so a new hire can start with less friction.
Those examples point to the same underlying pattern: Claude Cowork is being used to turn loose inputs into workable outputs. That does not mean the agent replaces judgment. It means users appear to be handing it tasks where the first version, format, or synthesis is the bottleneck.
Anthropic gives examples across roles. A lawyer can have documents formatted and filed, leaving more room for legal analysis. A hiring manager can use it to schedule interviews and summarize feedback, shifting more attention to candidates and work samples. A team lead can have slides prepared to explain a difficult decision, freeing time for the decision itself.
In each case, the agent is not presented as doing the core professional work. It is handling the surrounding material that makes the core work easier to complete, explain, or move forward.
Software work is a smaller share
Software development accounts for 8.7 percent of Claude Cowork usage in Anthropic’s breakdown. DevOps and infrastructure accounts for 7 percent. Those figures are meaningful, but they are much smaller than business operations and content creation.
Anthropic’s explanation is that interface matters. Developers, according to the company, tend to use Claude Code when they need to build, debug, and ship software. In Claude Cowork, they are more likely to handle the communication and coordination that surrounds technical work.
This distinction is important because both tools are tied to AI-assisted work, but they invite different behavior. Claude Code is specialized for programming. Claude Cowork brings agent-like capabilities into a familiar chat interface.
The source article notes that after Claude Code launched in 2025, some non-technical users used the terminal for tasks such as organizing folders, removing duplicate files, and writing spreadsheet formulas. For others, the terminal remained inaccessible. Cowork was designed to make similar capabilities available through chat instead.
That helps explain why coding is not the headline category here. The same underlying model capabilities can show up differently depending on whether the user is working in a terminal, a desktop app, a browser, or a smartphone.
The rest of the breakdown is fragmented
After the top two categories, usage spreads across several smaller buckets. Anthropic lists research at 6.4 percent, data analysis at 5.8 percent, document processing at 4.1 percent, sales operations at 4 percent, personal assistance at 3.8 percent, education at 2.4 percent, and meeting analysis at 1.8 percent.
These categories show that Claude Cowork is not limited to one department or one type of worker. But they also show that the broadest demand is for help with repeatable coordination, communication, and formatting work.
The taxonomy has limits. Anthropic notes that Business Ops classifies activities rather than professions. It also does not separate marketing, finance, or HR into their own categories. Those functions are likely included inside Business Operations, which Anthropic says probably helps explain why that category reaches a third of usage.
The study also has sampling constraints. It captures only a fixed maximum number of sessions per hour, which means peak periods are underrepresented. The numbers are percentages, not absolute volumes. About five percent of sessions involve personal use, such as hobbies or chatting, so the dataset is not purely workplace activity.
What this says about workplace AI
The analysis points to a grounded view of AI agents in organizations. The biggest immediate use may not be replacing specialist labor. It may be absorbing the recurring administrative load that sits around specialist labor.
That has implications for how companies evaluate tools like Claude Cowork. The value may show up in fewer blank screens, faster first drafts, cleaner handoffs, and better-organized internal information. Those gains are less dramatic than a fully automated workflow, but they are closer to the way many office jobs actually function.
Anthropic has recently expanded Cowork from the desktop app to the web and smartphones. Earlier, the company had given the agent the ability to directly operate Mac and Windows desktops. Those changes place the tool closer to the everyday environments where administrative and text-based work happens.
The source also places Claude Cowork and Claude Code alongside OpenAI's Codex, also known as ChatGPT Work, as examples of a broader shift discussed in a review paper by Meta, Stanford, and the University of Illinois. In that shift, code is not only an output. It becomes an operational layer for AI agents.
For now, Anthropic’s usage data suggests a simpler takeaway: the everyday office is full of work that needs structure, synthesis, and a first draft. Claude Cowork’s largest role is helping people move that work along.