TL;DR: To learn how to use Claude Cowork, start with a paid Claude plan (Pro, Max, Team or Enterprise), open Claude on desktop, web or mobile, select Cowork in the message box (or just ask, if you have the newer unified Claude), and describe the finished result you want. Point it at a folder or Project, review the plan, pick a permission mode, and let it run. Use /schedule for recurring work. Keep it in “Manually approve” mode for anything that touches money, messages or important files.
Claude Cowork is Anthropic’s agent mode for everyday knowledge work. Instead of answering one prompt at a time, it plans a multi-step task, works through it, and hands back real files such as spreadsheets, slide decks or reports. Anthropic’s help center describes it as bringing “Claude Code’s agentic capabilities to knowledge work beyond coding.” This guide explains how to use Claude Cowork step by step, based on Anthropic’s own documentation, plus the habits that make the output better.
What Claude Cowork is (and isn’t)
If regular chat is a conversation, Cowork is delegation. You describe an outcome, and Claude analyzes the request, makes a plan, breaks it into subtasks, runs code in an isolated environment on Anthropic’s servers, and delivers the output to your session. Anthropic says sessions run in the cloud (in beta on Team and Enterprise plans), so work keeps going if you close your laptop. If agents are new to you, our explainer on what agentic AI is covers the basics.
One recent change matters: Anthropic says “Claude Cowork is now just Claude” for some users. That unified experience is rolling out gradually to Pro and Max and is in beta for Enterprise. If your message box no longer shows separate “Chat” and “Cowork” options, you already have it, and Claude decides on its own whether your request needs a quick answer or a full task.
Where you can use Cowork
| Surface | Availability (per Anthropic) |
|---|---|
| Claude Desktop (macOS, Windows) | All paid plans; Windows needs the latest app version |
| Web (claude.ai) | Pro, Max, Team; Enterprise where an admin enables it |
| Mobile (iOS, Android) | Pro, Max, Team; Enterprise where enabled |
| Claude in Chrome side panel | Max and Team, rolling out to Pro; Enterprise where enabled |
The free plan does not include Cowork. For local files, browser use or computer use, you need the desktop app open and connected, even though the session itself runs in the cloud.
How to use Claude Cowork: setup in 4 steps
- Open Claude. Use the desktop app if the work involves files on your computer, or claude.ai and the mobile app for cloud-only tasks.
- Run guided setup. Anthropic’s Claude Academy tutorial recommends typing
/setup-claudein a new conversation. It walks you through picking your role, installing a matching plugin, and connecting tools such as Slack, Google Workspace or Microsoft 365. - Give it a place to work. Select Work in a folder to pick a local folder, or use a Project, which keeps its own files, instructions and memory across sessions.
- Describe the result. Select Cowork (if you see the toggle), describe the deliverable, review Claude’s approach, then let it run.

Choose a permission mode
Cowork has three modes that control when Claude stops to ask before acting, especially with connectors:
- Manually approve: Claude pauses and asks before actions. Safest for anything consequential.
- Automatically approve: Claude reviews each action for safety risks such as data exfiltration or prompt injection and blocks what it judges unsafe. Anthropic notes this mode uses more of your usage limit.
- Skip all approvals: nothing is checked. Anthropic says to use it only when you completely trust every action, connector and file involved.
Anthropic’s own advice is blunt: “For work with real consequences—money, messages sent as you, important files—stay close.” Cowork also asks for explicit permission before permanently deleting files. Our guide to what not to share with AI chatbots applies here too, perhaps more so, because an agent can act on what it reads.
Good first tasks
Anthropic suggests starting with something from your real to-do list that spans several tools or steps. Examples from its documentation include:
- “Organize my Downloads folder by type and date.”
- Turning receipts in a folder into a formatted expense report.
- Combining research notes into a one-page brief.
- Building an Excel file with working formulas, or a slide deck from meeting notes.
- Writing a status summary from last week’s Slack activity.
For more ideas on what agents do well, see our list of agentic AI workflow examples.
How to schedule recurring tasks
Scheduled tasks turn Cowork into a standing assistant. According to Anthropic’s scheduling guide, you can:
- Type
/schedulein any Cowork task, or click Scheduled in the left sidebar. - Choose New task, then Create with Claude (Claude asks questions and proposes the task) or Set up manually.
- For manual setup, enter a name, the prompt, the approval mode, a frequency (hourly, daily, weekly, weekdays or manual), and optionally a model and folder.
- Save. You can later pause, resume, edit, delete or run it on demand.
In the unified Claude experience, you simply describe the task and cadence in plain language, for example “Every Monday at 9 AM, summarize last week’s messages in my team’s Slack channels,” then confirm. Scheduled tasks run remotely, but Anthropic notes that tasks needing local files or apps only run locally.
Tips for better results
- Context beats clever prompts. In Anthropic’s best-practices article, the author argues that the gap between mediocre and great output is “almost never your prompt” but whether Claude has enough context.
- Make it ask questions first. The same article suggests adding: “Before we begin, repeat my ask back to me so we’re aligned, then ask me as many clarifying questions as you have.”
- Start with work you know well, so you can judge quality quickly.
- Set global instructions in Settings > Cowork for tone, format and role details.
- Watch usage. Multi-step tasks consume more of your limit; group related work into one task and check Settings > Usage.
Current limitations
Anthropic lists a few: sessions cannot be shared (individual artifacts can), some plugins with local MCP servers work only through the desktop app, and on Team and Enterprise plans admins control access and may require approval for write-capable connectors. Cowork is powerful, but it is still an agent with internet access, so treat its output like work from a fast new colleague: useful, and worth checking.



