Anthropic on Tuesday announced that Claude Cowork is now available on mobile and web, starting in beta for Max subscribers before rolling out to additional plans. The update converts Cowork from a desktop-only agent into a cross-device platform where tasks can be started on a laptop, continue autonomously in the background, and be reviewed from a phone after the user has closed the app.
The timing accompanies a usage dataset the company also released: 1.2 million anonymized Claude Cowork sessions sampled between May 11 and May 31, drawn from more than 600,000 organizations. The data indicates that the majority of Cowork activity is unrelated to writing software.
What people use Cowork for
By a wide margin the largest single category in the sample was business process and operations—tasks like consolidating scattered updates into a single report, building onboarding checklists, and reconciling spreadsheets—accounting for 33.4% of sessions. Content creation and copywriting (drafts, slide decks, posts, proposals) was the second-largest category at 16.4%.
Together those two categories represent roughly half of all Cowork usage in the sample. Software development accounted for just 8.7%. DevOps and infrastructure were 7%, research and intelligence 6.4%, data analysis and business intelligence 5.8%, document processing and extraction 4.1%, and sales and revenue operations 4%. The remaining 12 categories each made up under 4% of usage; for example, personal assistance was 3.8%, education 2.4%, and meeting intelligence 1.8%.
Anthropic frames these dominant use cases as “the work around the work”: tasks that span nearly every role in an organization but seldom appear in anyone’s core job description. Examples include drafting a status update, building a slide deck, or condensing reams of research into a single report.
What the mobile and web expansion changes
The rollout adds three concrete capabilities intended to fit Cowork into daily workflows:
- Session sync across devices: a task can be started on desktop, monitored from a phone, and the finished output retrieved from any device.
- Background, offline execution: Cowork can now run scheduled tasks even when the user’s device is not online. Anthropic gives an example of scheduling Monday morning client prep for 6 a.m., where Claude reads email threads, transcripts, and recent news, builds a briefing document, and leaves a follow-up email drafted but unsent for the user to review.
- Human-in-the-loop prompts surfaced to the phone: when Claude encounters a decision requiring judgment, it asks the user; Anthropic emphasizes that nothing ships without review and approval.
Desktop remains the most fully featured surface, with local file and browser access. The web version opens Cowork to users who cannot install a desktop application, which matters in enterprise settings where IT controls software installation. Anthropic also unified the interface: on web and desktop, chat and Cowork now share a single home screen and projects and artifacts persist across modes.
To encourage adoption, Anthropic is doubling Cowork usage limits through August 5, 2024.
Strategic logic: targeting non-developers
The mobile launch and the usage breakdown together show a two-track strategy. Claude Code, the terminal-based coding agent, dominates among software developers. Cowork is designed to capture a much larger population of knowledge workers who create, organize, and communicate information rather than write code.
Anthropic notes that Claude Code is primarily used by developers for the core parts of their role: building, debugging, and shipping code. When developers use Cowork, they tend to apply it to communications-focused work around their role—status updates, documentation, and coordination—rather than programming tasks.
This pattern matters for Anthropic’s business model. Developer-focused tools serve a narrower market. The Ramp AI Index in May indicated Anthropic had pulled ahead of OpenAI in business adoption for the first time—34.4% of firms paying for Anthropic’s services versus OpenAI’s 32.3%—with Claude Code identified as the primary driver. Cowork, however, targets an addressable market orders of magnitude larger: every knowledge worker who deals with a laptop, spreadsheets, and a slide deck due by Friday.
Competition and recent product updates
The mobile launch comes amid a flurry of product activity. Last week Anthropic introduced Claude Sonnet 5, a model that narrows the performance gap with more expensive Opus-class models while keeping lower pricing. Sonnet 5 is offered at an introductory price of $2 per million input tokens through August 31, 2024, after which it rises to $3 per million input tokens. Sonnet 5 powers Cowork and its improved agentic capabilities—better reasoning, tool use, and sustained task completion—directly enhance Cowork’s handling of complex, multi-step workflows.
Two weeks earlier Anthropic released Claude Tag, a Slack-native AI agent for team collaboration. Where Cowork focuses on individual task delegation, Claude Tag operates as a multiplayer tool: a single Claude identity everyone in a Slack channel can interact with, building context from conversations over time. Anthropic says 65% of the product team’s code is created by its internal Claude Tag.
Together Cowork and Claude Tag form a pincer strategy: Cowork captures individual cross-device productivity workflows, while Claude Tag embeds AI into team communication channels.
Security questions
The expansion arrives against unresolved security concerns. On July 1, 2024, security firm Armadin—led by Mandiant founder Kevin Mandia—published research describing what it called a full sandbox escape in Claude Cowork on Windows, involving DLL sideloading against the Claude desktop executable to gain access to Cowork’s virtual machine service and exploiting undocumented parameters to achieve root access and bypass network restrictions. Anthropic responded that exploiting the issue requires an attacker to already have local code execution on the host machine; Armadin warned that local VMs on nontechnical users’ systems create visibility gaps that endpoint security products struggle to monitor.
Moving Cowork to web and mobile changes the attack surface: those versions run tasks server-side rather than in a local VM, eliminating the specific local-VM risk Armadin described but raising different data-handling questions—especially for scheduled background tasks that process email threads, calendar data, and documents without real-time user oversight. Anthropic states that decisions still come to the user and nothing ships without review, but as Cowork handles more complex autonomous workflows, the surface area for prompt injection and data exposure grows.
When Cowork first launched in January, Anthropic explicitly warned about prompt injection risks, noting that these risks are not new but may be more visible when using an advanced tool that moves beyond simple conversation.
Geopolitics and infrastructure bets
Anthropic’s enterprise push also collides with geopolitics. CNBC reported that Alibaba will ban employees from using Anthropic’s AI tools starting July 10, 2024, placing Claude Code on a high-risk software list; this followed Anthropic’s June letter to the U.S. Senate accusing Alibaba of conducting what it called “the largest known distillation attack” against its models. Anthropic has also reportedly been closing loopholes that allowed Chinese companies to access Claude through third-country entities.
At the same time, Anthropic is investing heavily in infrastructure. Reuters reported that Anthropic signed a $19 billion, 20-year lease with TeraWulf for a data center being built in Hawesville, Kentucky, expected to provide 401 megawatts of computing power and become fully operational by 2028. Such a capital commitment suggests the company expects enterprise demand—not only from developers but from the millions of knowledge workers Cowork targets—to grow substantially.
Limits of Anthropic’s usage report
Anthropic acknowledges limitations in its usage analysis. The taxonomy classifies sessions by type of work performed, not by the job title of the person doing it. There are no standalone categories for marketing, finance, or HR, which are likely absorbed into the broad “business process and operations” bucket and may partly explain that category’s size. The sample is rate-capped rather than proportional to traffic, so the numbers reflect shares of sampled sessions rather than absolute volumes, and peak-hour usage is somewhat underrepresented. About 5% of sampled sessions involved personal, non-work use (hobbies, personal assistance, companionship-style conversations), so the data are not exclusively workplace activity. Anthropic also noted that its labeling pipeline changed around May 11, hence the choice of that start date for analysis.
What Cowork’s rise indicates about enterprise AI
The mobile and web launch and the usage statistics suggest enterprise AI is moving from proof of concept to proof of value. The central question for companies deploying AI is no longer whether the technology works but whether it delivers measurable productivity gains across organizations, not just within engineering teams.
Anthropic’s data indicate those gains may be emerging in an unglamorous but universal place: the connective, information-structuring work that moves organizations forward—status reports, onboarding checklists, variance memos, client decks—rather than the high-profile tasks of writing software or conducting research. By untethering that capability from the desktop and making it available on every device, Anthropic is betting that the most valuable AI agent may be the one that handles everything else.
