Every Claude launch this year has been about coding. Cowork isn't. It's Anthropic's answer to a different question: what does an agent look like when it's doing your job instead of your job's codebase?
As of July 2026, Cowork is expanding from a desktop-only feature to web and mobile, with background task execution that doesn't need a device online. Here's what it actually does, how the rollout works, and when it beats just opening a chat.
What Claude Cowork Actually Is
Cowork is an agent mode inside Claude that works across your connected tools — files, calendar, email, messaging, web browsers — to complete multi-step tasks, not just answer a single question. The framing matters: it's built for knowledge work, not development. Anthropic has said more than 90% of Cowork usage isn't software development at all — the largest categories are business operations and content creation.
That's the opposite profile of Claude Code, which is almost entirely developers. If Claude Code is "an engineer inside your terminal," Cowork is closer to "an operator inside your inbox and calendar."
What's New: Web and Mobile
Cowork launched desktop-first. The July 2026 update brings it to claude.ai in the browser and to native iOS and Android apps, with three concrete changes:
Sessions sync across devices
Start a task on desktop, check progress on your phone, pick up the finished output on the web — same session, same files, no re-explaining the task. Chat and Cowork now share one home tab on web and desktop: one sidebar, one search, one place for Projects and Artifacts.
Background execution with no device online
This is the actual unlock. Scheduled tasks now run without any device connected. Anthropic's example: set a Monday 6am client-prep task, and Claude works through email threads, transcripts, and recent news, builds the briefing doc, and leaves a follow-up email drafted — unsent, waiting for review — before you're even awake.
Human-in-the-loop, pushed to your phone
When Claude hits a decision only a person can make, it doesn't guess. It surfaces the question on your phone mid-task and waits. Nothing publishes, sends, or finalizes without your review.
How It's Rolling Out
Mobile and web access to Cowork started as a beta with a gradual rollout, beginning with Max plan subscribers and expanding to other paid plans over the following weeks. Desktop remains positioned as the deepest experience — it keeps access to local files and browser integration that web and mobile don't have yet. Anthropic also extended doubled Cowork usage limits through August 5, 2026, which suggests they're actively pushing adoption during the rollout window rather than gating it hard.
Access points: claude.ai on web, the Claude mobile app sidebar, or the Claude desktop app for the full feature set.
Cowork vs. Claude Code vs. Regular Chat
These three modes look similar from the outside — you type a prompt, Claude does work — but they solve different problems:
- Regular chat: single-turn or short back-and-forth, no persistent task, no background execution. Best for questions, drafts, and one-off analysis you'll review immediately.
- Claude Code: a coding agent that operates on a codebase through a terminal or IDE, with file edits, tool calls, and (in headless mode) CI/CD scriptability. Built for developers.
- Cowork: a longer-running agent across non-code tools — calendar, email, docs, messaging — that can run on a schedule, continue without a device online, and hand off between devices. Built for operational and content work.
If your task is "write this function" or "explain this error," that's chat or Claude Code. If your task is "every Monday, pull last week's client emails and drafts a briefing," that's Cowork — it's the only one of the three designed to run unattended on a recurring schedule.
Where Cowork Actually Fits
Based on Anthropic's own usage breakdown, the two biggest categories are business operations and content creation — things like status reports, meeting prep, research summaries, and first-draft content that gets human review before it ships. It's not positioned as a replacement for judgment calls; the human-in-the-loop design is explicit about that. It's positioned as removing the busywork that sits between "I have an idea" and "this is ready for someone to review."
The background execution is the feature that changes behavior, not the mobile app itself. A task that runs at 6am and hands you a finished draft by the time you check your phone is a different value proposition than a chatbot you have to actively drive turn by turn.
The Bottom Line
Cowork's web/mobile expansion isn't a UI refresh — it's Anthropic decoupling agent execution from having a device open. That's the same shift Claude Code went through with headless mode and scheduled automation, just applied to non-technical work. If you've been treating Claude as a chat window you have to babysit, this is the point where that stops being necessary for recurring, well-scoped tasks.