Anthropic’s Cowork report, released this weekend, upends the dominant narrative around Claude at work. Out of 1.2 million enterprise sessions analyzed between May 11 and May 31, 2026, across more than 600,000 organizations, software development accounts for only 8.7%. Business operations take 33.4%, and content creation another 16.4%.
Key Takeaways
- A third of Claude’s enterprise sessions handle business ops tasks, eight times more than software development.
- Anthropic frames the bulk of the volume as connective work: structuring and moving information, not producing code or finished deliverables.
- The 33/8 ratio reshapes how vendors talk about Claude in the market and how buyers should allocate their seats internally.
What Claude Actually Does Behind the Office Door
The Cowork report is the first large-scale snapshot of what enterprises actually do with Claude. Anthropic aggregated 1.2 million anonymized sessions, cross-referenced more than 600,000 organizations, and tagged each conversation with the task performed. Three blocks stand out.
The first, business ops, soaks up 33.4% of the volume. In practice: status reports, meeting recaps, board slides, coordination emails, project briefs, budgets to reshape. Nothing heroic. Nothing like the code-generation demos labs have been pushing for eighteen months.
The second block, content creation, holds 16.4%. Newsletters, marketing copy, internal documentation, blog notes, sales talking points. Still office work, still text to reshape and route.
Software development stops at 8.7%. Roughly four times less than business ops. Anthropic does not hide the gap between that share and Claude’s public image, which has been shaped by Claude Code’s momentum among developers. Devs are still a visible, high-volume audience, but they are no longer the majority of enterprise usage. Our read: the coding demo sells the tool, the connective task keeps it running.
The rest of the volume splits between data analysis, research and monitoring, internal support and HR questions. Nothing exotic, everything that fills an average team’s day. A meaningful share of sessions touches billing questions tied to Anthropic’s recent shift to token-based API pricing on Fable 5, which now exposes the enterprise to the true shape of its usage. Keeping that bill in check matters, learning to save on tokens.
Why Anthropic Publishes These Numbers Now
The timing is not innocent. Anthropic is playing two battles in 2026. The narrative battle first, against an OpenAI that still owns the mental model of consumer AI. The enterprise battle next, against a Microsoft Copilot pushing hard and a Google threading Gemini into Workspace.
Publishing the 33/8 ratio lets Anthropic reframe its position. We see three implicit messages. First: Claude is already inside the teams that matter, not just among developers. Second: CIOs signing today are signing for augmented office work, not for the AGI roadmap. Third: the value edge sits on the quality of connective tasks, not on raw code benchmarks.
This comes in the wake of a funding round aiming at a 900 billion dollar valuation for Anthropic. Investors look at enterprise traction, not just tokens billed. A solid number on business ops adoption reassures them on corporate customer stickiness.
Anthropic also unpacked what it calls connective tasks. This is not about producing a standalone deliverable. It is about pulling, structuring and passing information between teams, between hierarchy layers, between tools. The PDF summary for a director, the recap of last week’s decisions, the translation of a technical spec into legal-friendly wording: none of that shows up in public benchmarks, yet it fills entire workdays.
More articles on Horizon
- Apple Sues OpenAI Over iPhone Secrets and 400 Hires
- Muse Spark 1.1: Meta’s New AI Model Beats on Price
- Grok 4.5 Test: We Rate Musk’s Coding Promise
What This Report Forces Rivals and Teams to Change
On the competitive side, the release exposes rivals’ blind spots. OpenAI and Google communicate almost exclusively on agentic capability and reasoning benchmarks, two dimensions largely absent from the real enterprise usage mix Anthropic just published. Those labs will either need to release a comparable report to reframe the reading, or accept that the industry conversation is now anchored on Anthropic’s home turf.
Microsoft Copilot sits in an awkward spot. Office productivity is its historical territory, but its engine still leans heavily on GPT. If the narrative “Claude is the connective task leader” sticks, Microsoft has to accelerate deeper Claude integration in Office workflows, even if that adds tension to its OpenAI partnership. Google, already integrated into Workspace, struggles to get Gemini adopted at the level of daily admin tasks.
For buyers already sitting on Claude licenses, the fallout is immediate. Three choices stack up. One: reallocate developer seats toward ops, marketing and HR teams where the actual value shows up. Two: rewrite internal prompting guides to cover the real frequent cases, not the polished code demos. Three: measure the internal ratio directly, so procurement policy tracks true usage rather than vendor storytelling.
Development teams do not lose their tool, but they lose their representation monopoly. That will not change their immediate productivity. It will weigh, however, on next year’s budget cycle: when a CIO sees 90% of seats sit under ops usage, the volume conversation with Anthropic shifts. Purchasing moves from “how many developers do I equip” to “how many teams will use it daily”.
For executives running AI transformation, the report is a useful reminder: the value of an LLM in the enterprise is read in Monday morning emails, not in Friday night demos. Claude Code’s commercial momentum at Anthropic remains real, but it no longer sets the production line. The production line, right now, is the status report nobody wants to write.
Follow the story on Horizon.
