Anthropic launches Claude Cowork on web and mobile for Max subscribers, turning a desktop-only agent into a cross-device work companion.
Usage data from 1.2 million sessions shows the tool’s sweet spot: “work around the work” like report assembly, onboarding checklists, and spreadsheet reconciliation.
The move mirrors OpenAI’s Codex pivot — both labs now bet that owning the work surface matters more than owning the best chatbot.
Background execution without a device online becomes a new differentiator; the agent runs while your laptop sleeps.
Anthropic just made its administrative agent portable. Starting Tuesday, Claude Cowork — the Claude Code-style tool that launched on desktop in January — lands on web and mobile for Max subscribers. You can kick off a task at your desk, watch progress on your phone, and collect the finished output with your laptop still closed.
That shift signals intent. Anthropic no longer wants Cowork mistaken for a coding assistant for non-coders. It wants an agentic administrative coworker: background execution, cross-device presence, human-in-the-loop when a decision demands your judgment. The coding agent wars have officially spilled into the rest of the office.
OpenAI made the same calculation with Codex. What began as a developer tool now powers reports, spreadsheets, presentations, research, and data analysis for people who never write a line of code. Both labs have concluded that the next moat isn’t the model — it’s the surface where work actually happens. Chat interfaces are commodities. The persistent, context-aware workspace is the prize.
Anthropic’s parallel play, Claude Tag, embeds an always-on teammate inside Slack. That’s not a coincidence. The strategy is clear: colonize every surface where knowledge work lives. Desktop for deep file-and-browser access. Web and mobile for continuity. Slack for the conversational layer. Each entry point reinforces the others.
The multi-platform architecture unlocks a capability that single-device tools cannot match: background execution without a local machine online. Anthropic’s example is pointed: “Set Monday’s client prep for 6 am: Claude works through the email threads, transcripts, and recent news, builds the briefing doc, and leaves the follow-up email drafted but unsent. Review it over coffee.” The laptop stays shut. The agent runs in the cloud. The human reviews output, not process.
Desktop remains the power user tier. Local files, browser automation, deep integration — those stay anchored to the installed app. But web and mobile expand the addressable market to anyone with a browser. Anthropic says chat and Cowork will unify on web and desktop first, with projects and artifacts shared across both. The product line is collapsing into a single workspace.
The usage data Anthropic released is the most revealing part of this announcement. The company sampled 1.2 million anonymized Cowork sessions across 600,000 organizations in the last two weeks of May. The dominant use case — 33.4 percent — is what Anthropic calls “business process operating.” Pulling scattered updates into a single report. Building onboarding checklists. Reconciling spreadsheets. Finance, HR, and administration roles drive this volume.
Next at 16.4 percent: content creation and copywriting. Drafts, slide decks, social posts, proposals. The pattern is unmistakable. Cowork excels at the “work around the work” — the tasks that keep companies functioning but rarely appear in anyone’s job description. These are the glue tasks. The handoff friction. The administrative drag that consumes hours without producing core value.
That insight should unsettle every SaaS vendor selling vertical workflow tools. If a general-purpose agent can assemble onboarding checklists and reconcile spreadsheets better than a purpose-built HR or finance module, the moat for specialized software shrinks. The agent doesn’t need domain-specific UIs. It needs context, tool access, and the ability to iterate until the output passes human review.
Skeptics will note the data covers only two weeks of Max subscribers — early adopters, likely tech-forward teams. The distribution may shift as broader cohorts arrive. But the shape of the demand is already visible. Knowledge workers don’t want a smarter chatbot. They want the drudgery handled while they focus on the decisions that require judgment.
Anthropic’s bet is that the winning agent will be the one that follows you across devices, runs while you sleep, and interrupts only when a choice exceeds its authority. The mobile and web launch is the first real proof that they’re building for that loop. The desktop app was a prototype. This is the product.
The rest of the office is watching. Microsoft has Copilot embedded across 365. Google has Gemini in Workspace. But those are add-ons to legacy suites. Anthropic and OpenAI are building native agent workspaces from scratch, unburdened by file formats or permission models designed for human collaboration. They move faster because they have less to protect.
For now, the advantage goes to the lab that makes background execution feel ordinary. Set the task. Close the laptop. Get the result. That simple loop — if it works reliably — changes how a day feels. The code wars are over. The administrative agent war has begun.