What Is Claude Cowork? Longer Jobs, Handed to Claude
Claude Cowork is Anthropic’s way of handing Claude a whole job instead of a single question: it plans the steps, works through them, and comes back with finished files. What it is, where it runs, what it can reach, how approvals and scheduled tasks work, and which plans include it.
8 min read
Claude Cowork is the part of Claude that takes on longer, multi-step jobs for you. Instead of answering one prompt at a time, Claude plans the work, breaks it into steps, runs them, and hands back finished files: a spreadsheet with working formulas, a slide deck, a formatted report, a sorted folder. It runs on Anthropic’s servers, reaches your own files and browser through the Claude Desktop app, uses the apps you have connected, and can run on a schedule. It is on paid plans only, and it is in the middle of becoming simply “Claude”: on some accounts there is no separate Cowork option any more, because every conversation can take on a task. Everything below reflects Anthropic’s help center as of September 30, 2026.
What Cowork is
Anthropic’s guide to getting started with Claude Cowork (opens in a new tab) says Cowork brings Claude Code’s agentic capabilities to knowledge work beyond coding, using the same agentic architecture with no terminal required. You describe an outcome, step away, and come back to finished work. The same page lists what that means in practice:
- Long-running tasks that are not cut short by conversation timeouts or context limits.
- Sub-agents: Claude splits complex work into smaller tasks and runs them in parallel.
- Professional outputs, such as Excel spreadsheets with working formulas, PowerPoint presentations and formatted documents, which you preview and download from the session.
- Projects that group related tasks with their own files, instructions and memory.
- Scheduled tasks that run on a cadence you choose.
- Visibility while it works: progress for each step, its reasoning, and the chance to steer it mid-task.
Cowork is becoming “just Claude”
The name is in flux, so it helps to know which version you have. Anthropic’s page Claude Cowork and chat are one Claude (opens in a new tab) describes a new experience, rolling out gradually from the Pro and Max plans on web, desktop and mobile, in which chat and Cowork merge into one conversation. You ask for what you need, and Claude decides whether that is a quick answer or a task. If your message box still shows separate “Chat” and “Cowork” options, you do not have it yet. Once you do, you cannot switch back, and your tasks, projects, connectors, skills and files come with you.
Where Cowork runs
Cowork sessions run in the cloud, in beta, in an isolated environment on Anthropic’s servers. Your sessions and files are saved to your Claude account, so you can start a task on one device, check on it from another, and pick up the result wherever you are. Work continues when you close your laptop.
- Claude Desktop for macOS and Windows: the full experience, and the only way for Cowork to reach things on your computer. Windows needs the latest version of the app.
- The web at claude.ai and the Claude mobile apps: in beta on Pro, Max and Team, and on Enterprise where an admin has turned it on.
- The Claude in Chrome side panel: opening the panel starts a Cowork session directly, on Max and Team, rolling out to Pro, and on Enterprise where enabled.
One change is dated. Anthropic’s page on Cowork on web, desktop and mobile (opens in a new tab) says that on October 6, 2026, new Cowork tasks on Pro and Max plans run in the cloud, and the “Only on your computer” option in Settings, General is removed. Tasks already started on your computer stay there until they are done. Your folders stay on your computer: Claude reaches only the ones you have connected, through the desktop app, while it is open, and fetches a copy of a single file when a cloud task needs it. For work that must stay on one machine, the same page points to Claude Code in the desktop app instead.
What it can reach
- Local files. On desktop, Claude reads and writes files in folders you have connected, with no manual uploads. It needs your explicit permission, with an “Allow” prompt, before permanently deleting anything.
- Connectors. Apps you have connected, such as Google Drive, Gmail, Microsoft 365 or Slack, work while Claude carries out a task. You choose which ones from the “+” menu in the message box or under Customize, Connectors. What each kind of connector can reach is in Claude connectors explained.
- The browser. For tasks that touch websites, Claude can open pages, read, click, type and fill in forms. On desktop it uses a browser built into the Claude Desktop app by default, or your own Chrome if you already use Claude in Chrome; see Claude in Chrome for that side.
- Your screen. Computer use, where Claude operates other apps directly, is in beta on Pro and Max plans and asks before using each app.
- Skills and plugins. A plugin bundles skills, connectors and sub-agents for a role, and one you add works in chat and Claude Code as well.
- The web. Web fetch runs on Anthropic’s side and is limited to search results and URLs you have shared.
Approvals and safety
A permission setting in the message box controls how often Claude stops to ask. The getting-started guide lists three modes. Manually approve pauses for your Allow or Deny on actions. Automatically approve keeps working while Claude reviews each action for safety, such as data exfiltration or prompt injection, and blocks anything it judges unsafe; it uses more of your usage limit because of that extra check. Skip all approvals has no pauses and no automatic check. Anthropic’s pages differ here: in the new merged experience, the one-Claude page lists only Auto and Manual, with Manual the default.
Anthropic’s page on using Cowork safely (opens in a new tab) is frank that isolation protects your computer and network from the code Claude runs, but does not limit what Claude can read or do through the access you grant. The main risk is prompt injection: instructions hidden in a web page, email or document that try to steer Claude. Its advice, in brief:
- Give Claude a dedicated working folder, and keep financial documents, credentials and personal records out of reach.
- Switch to Manually approve when a task touches sensitive files, accounts or sites, uses a tool for the first time, or would be hard to undo, such as sending messages or making purchases.
- Watch the task, not every command: is Claude opening files or sites you never mentioned, or drifting beyond what you asked?
- Be especially careful with computer use, which has no sandbox between Claude and your screen.
On Team and Enterprise plans an admin can turn off Automatically approve for the organization and require approval for write-capable connector tools.
Scheduled tasks
A scheduled task is a prompt Claude saves and runs on a cadence: hourly, daily, weekly, on weekdays, or on demand. Type /schedule in a Cowork task, click Scheduled in the left sidebar, or, in the new experience, just say “Every Monday at 9 AM, summarize last week’s messages in my team’s Slack channels” and confirm what Claude proposes. Each run is its own Cowork session, with the same connectors, skills and plugins as any other task.
Anthropic’s page on scheduling recurring tasks (opens in a new tab) says they run remotely, even when your computer is asleep or the desktop app is closed, and work with your connectors and the files saved to your Claude account rather than a folder on your computer. Its manual setup form, though, has an optional working folder and notes that a task needing local files or apps runs only locally, and the October 6 notice says scheduled tasks that use files on your computer move to the cloud and need the desktop app open. If a task depends on local files, test it with the app closed before you rely on it. For anything that runs while you are away, Anthropic advises starting with low-risk summaries and reviewing each run.
Plans, by name
Cowork is on paid plans only: Pro, Max, Team and Enterprise. On desktop it is on every paid plan; on the web and mobile, Enterprise has it only where an admin has enabled it. There is no separate allowance: longer tasks that search the web, run code or create files draw more from your plan’s usage limits than a quick question does, and you can check what you have used under Settings, Usage.
How it differs from chat and from Claude Code
- Chat answers you in the conversation and waits for your next message. Cowork takes an outcome, runs unattended, and hands back files. In the merged experience the difference is in the request, not a mode you pick.
- Claude Code works in a code repository and ends in a diff or a pull request. Cowork ends in a document, a spreadsheet, a deck or a sorted folder. The full comparison, with five jobs sorted between them, is in Claude Cowork vs Claude Code.
- Projects work in both; see Claude projects in Cowork.
How to use Claude Cowork for the first time
- Open Claude Desktop, the web app or the mobile app on a paid plan. In the message box, select “Cowork”, or, if there is no such option, stay in any conversation.
- If the job needs local files, connect a folder in the desktop app. Pick a working folder, not your whole drive.
- Set the permission mode to Manually approve for the first few tasks.
- Describe the outcome, the inputs and the format, and say what Claude must not do.
- Review Claude’s plan, let it run, and check the files it hands back before you use them.
In my connected folder "Q3 Invoices", read every invoice and build a spreadsheet: date, vendor, category, amount, file name. Add a second tab with totals by category. Do not rename, move or delete any file. List any invoice you could not read.
Fifteen more jobs with prompts to adapt are in Claude Cowork examples. If Cowork will not start or does not show up, especially on Windows, see Claude Cowork not working.
Where the results go
Anthropic lists one limit that matters for teams: Cowork sessions cannot be shared, although artifacts made in them can. What a task produced, and what it could not do, stays in your session list. fenbs is a task board that Claude connects to as a custom connector over MCP, at https://fenbs.ai/api/mcp, signing in through your browser with the scopes you tick. A Cowork task can read its task on the board first, then comment what it produced when it finishes, and History records the change under Claude’s name. Leave the move to Completed to a person who has checked the file. fenbs does not store the files themselves and has no due dates or assignee field, so the note says where the output lives and who owns it.
Related
Connect Claude to a board: Claude and fenbs and how to connect Claude to your project board with MCP. Claude across a whole project: Claude for project management. OpenAI’s equivalent: ChatGPT Work vs Claude Cowork.