Anthropic just published a big report on how people actually use Claude at work, and the answer is surprising for anyone who thought AI was mostly a coding tool. Out of 1.2 million work sessions studied in May 2026, only 8.7% were about writing code, while 33.4% were about everyday office tasks like status reports, slides and emails.
The Gist
- Claude spends most of its time doing boring office tasks, not writing software.
- Business ops (reports, slides, meeting recaps) eat up a third of all sessions.
- Only about one session in twelve is dedicated to coding.
- Anthropic calls the biggest use case “connective work”: pulling info together and passing it around.
Meet the Cowork Report, Anthropic’s First Big Look Inside the Office
Anthropic is the company behind Claude, one of the two or three AI chats a beginner is likely to have heard of, alongside ChatGPT and Gemini. Until now, most of what we knew about Claude at work came from marketing demos and developer stories. The Cowork report changes that. It is the first time Anthropic pulls back the curtain on what people really do with Claude when nobody is watching.
The setup is straightforward. Anthropic looked at 1.2 million work sessions from more than 600,000 organizations, all anonymized. Then it sorted every session into a bucket, based on the kind of task Claude was helping with. That gives us a shape of the real usage, not a highlight reel.
And the shape is nothing like the picture you get from tech Twitter. The biggest bucket, at 33.4%, is business ops. That means status reports, meeting notes turned into action items, presentations, emails, project briefs. Nothing flashy. Just office work, faster.
The next big bucket is content creation at 16.4%. Newsletters, product descriptions, blog posts, internal documentation. Coding, the use case everyone talks about, sits way lower at 8.7%.
What “Connective Work” Actually Means for a Beginner
Anthropic has a specific name for the biggest slice of usage: connective work. If you have never heard the term, that is fine. Almost nobody had, until this report.
Connective work is not about producing a shiny finished thing. It is about the invisible glue: taking information from one place and making it fit somewhere else. Turning last week’s meeting into an email a director can read in ten seconds. Reshaping a legal brief into plain-English bullet points for the sales team. Summarizing a project’s Slack chatter into a Monday morning update.
If you have ever spent a Sunday evening cleaning up notes for Monday’s team update, you have been doing connective work by hand. What Anthropic just proved is that when a company gives its people access to Claude, most of them use it exactly there. Not for coding. For the boring glue.
This matters for you, the beginner, because it also tells you where to start. If your first attempt at using Claude was “write me an app” and you gave up, you were trying the smallest bucket of usage. If you start instead with “turn my meeting notes into a two-line action list”, you are trying the biggest bucket, the one the rest of the market has already validated for you.
Keep learning on AI Noobies:
- Claude Fable 5 Wrote 1M Lines of Code in 11 Days
- Muse Spark: Meta’s New AI Model That Uses Your Apps
- Claude Reflect Shows How You Actually Use AI
Why the “AI Equals Coding” Story Was Never the Whole Picture
For two years, the loudest AI story has been the one about coding. Big models generating apps, junior developers replaced overnight, GitHub copilots writing entire pull requests. That story sells subscriptions and drives headlines. It does not describe the office life of a real 600,000 organizations sample.
Two things happened at once. First, the coding demos are easy to film. It looks impressive on video when a chatbot spits out a working game in thirty seconds. Second, developers were the earliest paying customers because Anthropic’s tool Claude Code and its rivals landed on their desks first.
But once the tool spread beyond engineering, it followed the shape of the actual workforce. Every company has ten to twenty times more people writing meeting recaps than writing Python. So the usage naturally piled up on office tasks, not on code. Anthropic is now the first to say it out loud with a number attached.
The wider public conversation is about to shift too. Journalists, HR consultants, unions and school teachers have all been reading the same “AI will replace software engineers” script. This report gives them a different one: AI is quietly settling into the desks of people who make PowerPoints, write to their colleagues, and rewrite last week’s notes. That is a much bigger group than developers.
How to Try Claude on Real Office Work This Week
A beginner walking away from this news has one clear next step. Skip the “make me an app” experiment. Try the connective work bucket, because that is where the market has already voted with its usage.
Open Claude in a browser at claude.ai. Paste any long email thread you received this week, and ask “give me a two-sentence summary and one clear action for me”. This is a classic connective task. You will see the tool do exactly what the 33% of enterprise sessions do every day.
Then try a second experiment. Take the last meeting note you have, no matter how messy. Ask Claude to rewrite it as a status update for a busy manager who has three minutes. This is the second most common use, and again, no code required.
The free plan is enough for these first two experiments. You will run into daily limits within a few days if you use it seriously, and at that point the twenty-dollar Pro plan starts making sense. That said, you do not need to pay to run these first two experiments. Try them once, and you already have the same core reflex the 600,000 companies from Anthropic’s report have built into their teams. For a deeper look at how Anthropic frames its own tool in workflows, Horizon covered the recent switch to token-based pricing on Fable 5, which changes how enterprises get billed for exactly this kind of usage.
Stay tuned on AI Noobies.
About the Author
