If you've been searching "claude cowork" and wondering whether it's for you, the short answer is that it's the part of Claude built for handing over a whole task. In a normal chat you ask, it answers, you ask again. With Cowork you describe the outcome you want, Claude makes a plan, works through the steps and comes back with something finished: a formatted document, a spreadsheet with working formulas, or a presentation you can open in PowerPoint.

You don't need any technical background. You describe the job the way you'd brief a capable colleague.

Who can use it

Cowork comes with paid Claude plans: Pro, Max, Ultra, Team and Enterprise. It isn't on the free plan. There's no separate Cowork fee listed, but it counts toward your plan's usage, and a multi-step job uses more of that than a quick question. You can check where you stand at Settings > Usage.

On desktop it's available on all paid plans. On web and mobile it's available on Pro, Max, Ultra and Team (in beta on Team), and on Enterprise when an owner switches it on. At work, ask your administrator before assuming it's allowed.

You might not see a Cowork button

This trips people up. Anthropic is merging chat and Cowork into one experience, rolling out gradually to Pro, Max and Ultra, and on Team and Enterprise wherever it's been turned on. If your message box shows "Chat" and "Cowork" options, you're on the older layout: select Cowork. If those options are gone, you're on the new one, and any conversation can take on a Cowork task. Nothing is missing. Just describe the job.

Start your first task

Open Claude on the web at claude.ai, in the desktop app or in the mobile app. Select Cowork if you see the option, then describe the task. Claude shows its approach. Read it, correct anything that's off, and let it run.

The brief matters more than it does in chat, because Claude will go further on its own before you see anything. Include the outcome, the source material, the audience, and the point where it should stop and check with you.

I've attached five customer interview notes from last quarter. Produce a two-page Word summary for my manager, who hasn't read them. Structure: the top three themes, one short quote for each, and what we should do next. Use only what's in these notes. If something is unclear or the notes disagree, list it at the end rather than guessing. Show me your plan before you start.

Sessions run in the cloud, so a task keeps going if you close your laptop. The exception is anything on your own computer: if a task uses local files or your browser, keep the Claude desktop app open so Claude can reach them.

Pick your approval setting before you start

This is the most important choice. There are up to three modes, and the names have changed recently, so yours may look slightly different.

Manual. Claude pauses and asks before actions. This is the one to use while you're learning, and in the new Claude experience it's the default on Enterprise.

Auto. Claude keeps working and reviews each action for safety, blocking anything it judges unsafe. In the new Claude experience this is the default on Pro, Max, Ultra and Team, so if you're on one of those plans, check it before you start. Anthropic itself says this doesn't replace your judgement, and suggests staying close or switching back to Manual when money, messages sent in your name or important files are involved.

Skip. Claude doesn't pause, and nothing checks its actions. Leave this one alone unless you completely trust every file, app and website involved in the task.

Separately, Claude has to ask for your permission before it permanently deletes any file.

What to keep out of it

Giving Claude real abilities creates real risks, and they depend on two things: what it can read and what it's allowed to do. Three habits cover most of it.

Connect only what the task needs. Point it at one project folder, not your whole drive. Anything Claude opens is processed on Anthropic's servers rather than staying on your computer, so follow your workplace's rules on what can go into an AI tool.

Be careful with outside content. A web page or document can carry hidden instructions written to trick an AI into doing something you didn't ask for. Anthropic calls this prompt injection. If a task involves browsing or reading unknown files, use Manual and don't give it access to anything sensitive at the same time.

Check the output. A finished-looking file isn't a checked file. Our guide to verifying AI output before you send it has a routine you can use.

๐Ÿ’ก Cowork or chat?

Use chat for a quick question or a short draft. Use Cowork when the job has several steps and ends in a file, such as turning a folder of notes into a report. If you'd normally block out an hour for it, it's worth trying as a Cowork task.

Try it this week

Pick one low-stakes job you already know how to do well, such as turning last month's status notes into a one-page summary. Set approvals to Manual, paste in the brief above with your own details, and time how long it takes to check and fix the result. Compare that with doing it yourself, and you'll know whether Cowork belongs in your week.