Claude Cowork reaches web and mobile, pointing the coding-agent pattern at office work

Anthropic moved Claude Cowork off the desktop on July 7, 2026, bringing it to web and mobile beginning with the Max plan. Cowork launched as a desktop app in January as a Claude Code-style agent aimed at general knowledge work rather than code, and the release that matters here is less the new platforms than what runs underneath them: sessions now execute remotely, so a task started at a desk can finish while the laptop is closed. That reads as Anthropic generalizing the pattern it proved with coding agents to the wider category of office work.

What shipped

Per Anthropic’s release notes, Cowork is now available on web and mobile in addition to desktop, rolling out first to Max subscribers with more plans to follow. The mobile client is in beta. The consequential change sits behind the platform list: Cowork now runs sessions remotely, also in beta, with sessions and files saved to the Claude account rather than tied to a local machine. Work continues after the laptop closes, and scheduled tasks run with no device online. Anthropic also merged the surfaces, so Chat and Cowork now share one home with a single place for projects and artifacts across both.

The product itself is positioned as an agent that takes actions rather than returns answers: compiling research into a deck or spreadsheet, drafting content, building onboarding checklists, and running multi-step business processes. Anthropic’s framing points it at finance, HR, administration, marketing, and management, the roles doing the work around the work. TechCrunch, citing Anthropic usage data from roughly 1.2 million anonymized sessions across more than 600,000 organizations in May, reported that business-process work made up 33.4% of Cowork activity and content creation 16.4%, with software development at 8.7%, a distribution that explains why Anthropic is investing in a non-coding surface at all.

For a working professional, the practical shift is the same one the coding tools already made: the bottleneck stops being your presence at a machine. An agent that holds its own session, keeps running unattended, and reports progress to a phone turns the meeting hour and the commute into time work can advance, which is the actual value rather than doing spreadsheet work on a touchscreen.

Where this lands in the market

This is the coding-agent interaction model crossing into general office software, and the timing is not incidental. Cursor shipped an iOS app in late June built around the same premise, that supervising an autonomous agent should not require a laptop, and Anthropic’s own Claude Code has offered mobile reach for months. What Cowork does is take that pattern, start a long task, walk away, check it from anywhere, and aim it at the finance analyst and the marketing lead instead of the developer.

The remote-session piece is the part that changes the competitive frame. Once a Cowork task lives in the account rather than on the device, scheduled and unattended runs become the default expectation, and the comparison set widens from AI chat assistants to the broader automation and agent-platform market. It also raises the same question every unattended-agent product now carries: work that proceeds without anyone watching still needs review discipline, and the teams that benefit will be the ones whose oversight survives the convenience.

What’s worth watching

1. How fast it leaves the Max tier. Starting on Max, with mobile and remote sessions both in beta, leaves real coverage questions. How quickly Cowork reaches Pro and Team, and whether usage limits differ by surface, will decide how much actual work moves onto it.
2. What unattended runs cost. Scheduled tasks that run with no device online bill for compute whether or not anyone is watching the output. The durable question is what always-on, mobile-initiated agent usage does to a seat’s effective cost once the pattern becomes routine.
3. Whether general knowledge work rewards agents the way code does. Coding has crisp success signals: tests pass, the build is green, the diff merges. Business-process work is fuzzier to verify, and whether Cowork’s start-and-supervise loop holds up without those signals is the open question under the whole category.

The quieter shift is that the autonomous-agent interface, start a task, close the lid, approve the result from a phone, is escaping the developer tools that invented it. Anthropic is betting the office runs on the same loop as the codebase. Stackmaven’s follow-up coverage will revisit plan availability and adoption on or around October 6.

Sources cited

1. Anthropic: Claude apps release notes (July 7, 2026) support.claude.com
2. Claude: Cowork product page claude.com
3. TechCrunch: The coding agent wars are spilling into the rest of the office techcrunch.com