The coding agent hype was a head fake — Anthropic just published the receipts showing what people actually do with AI agents, and it's reconciling spreadsheets.
The Summary
- Anthropic launched Claude Cowork on mobile and web, expanding beyond desktop so tasks can start on your laptop, run in the background, and be reviewed from your phone after you close the app
- Usage data from 1.2 million sessions across 600,000+ organizations reveals 33.4% of Cowork use is business operations, not coding — the narrative around AI agents has been aimed at the wrong market
- Tasks now persist across devices and continue autonomously even when you're offline, turning agents from desktop tools into ambient infrastructure
The Signal
Anthropic released something more valuable than a feature update this week. They released data that rewrites the assumptions about who AI agents are actually for. Between May 11 and May 31, the company sampled 1.2 million anonymized Claude Cowork sessions. The largest category wasn't coding. It was business process and operations at 33.4%. People pulling scattered updates into reports. Building onboarding checklists. Reconciling spreadsheets.
The coding agent story dominated 2024 and 2025 because it was legible to the people funding AI companies. Developers are a known market with known budgets and known pain points. But the data suggests the real market is everyone else. The knowledge workers who don't write code but drown in the kind of tedious, multi-step work that agents handle better than humans ever wanted to.
"Your work goes everywhere with you, and keeps going without you."
This is the mobile and web expansion's actual value proposition. Cowork now runs cross-device, so you can start a task on desktop, close your laptop, and check the results from your phone hours later. The agent keeps working in the background even when you're not logged in. That's not a convenience feature. It's a redefinition of what an agent is. It stops being a tool you use and starts being infrastructure that runs beneath your workday.
The mobile push matters because phones are where non-technical users live. Developers tolerate clunky desktop interfaces. The rest of the workforce won't. Anthropic is rolling out the beta to Max subscribers first, with expansion to other plans coming later. The sequencing is smart. Early adopters stress-test the cross-device handoff. Broader rollout happens once the experience is smooth enough that someone's VP of Operations can actually use it.
Key capabilities now enabled:
- Task persistence across devices without manual syncing
- Background execution that continues when apps are closed
- Mobile review and approval of agent-generated work
- Cross-platform task initiation and monitoring
The usage data changes the competitive landscape. If the biggest use case is business operations, not software development, then the companies best positioned to win aren't the ones with the best code completion models. They're the ones that can embed agents into the workflows of people who've never used an API. Anthropic just published a roadmap for everyone else to follow, and simultaneously shipped the product that executes on it.
The Implication
If you're building AI tooling, this data should recalibrate your target user. The person reconciling spreadsheets has a bigger budget, more pain, and less patience for developer-first interfaces than the engineer who already uses GitHub Copilot. Design for them.
For knowledge workers, the background execution model is what to watch. The leap from "tool I open" to "infrastructure that runs" is the same leap email made from something you checked twice a day to something that was always on. Cowork mobile is Anthropic testing whether agents can make that jump. If they can, the next question is how many of your current tasks you'll even need to touch anymore.