How to Use Claude Cowork: A Getting-Started Walkthrough (2026)

TL;DR. Claude Cowork is the "do it for me" version of Claude, included in Claude Pro. To actually use it: (1) get Pro and open Cowork on desktop, (2) create a Project (this gives it a folder and memory), (3) open a task inside that Project and describe an outcome in plain English, (4) review and approve as it works. No code. The whole skill is learning to delegate clearly and to keep a checkpoint on anything irreversible.

If you've read what Claude Cowork is and want to actually start using it, this is the practical walkthrough. It takes about ten minutes to get going.

Step 1 — Get access

Claude Cowork is included in the Claude Pro plan (about $20/month) and the Max plans. There's no separate "Cowork" purchase. Subscribe at claude.ai, then open the Claude desktop app — Cowork lives there, because it works with your real files and apps, which a browser tab can't do. (The free plan doesn't include Cowork.)

Step 2 — Create a Project before you do anything else

This is the step people skip, and it's the most important one. A Project is a persistent workspace: it has its own folder, its own instructions, and its own memory. That's what lets Cowork remember your context instead of starting cold every time.

Open Cowork → Projects → create a new one. Name it for the work you'll do there ("My Client Work," "Marketing," "My Practice"). Cowork attaches a folder to it automatically. Anything that needs to stick around — files you're working on, a saved profile, instructions — lives in that folder.

Why it matters: tools and plugins that run outside a Project can't save state between tasks. If you ever install a plugin that asks you to "run the setup wizard" and it can't find a Project, this is why. Make the Project first.

Step 3 — Give it your first real task

Open a task inside your Project and describe an outcome, not a step. Good first tasks are small and low-stakes:

• "Here are my rough meeting notes — turn them into a clean summary with action items."
• "Organize the files in this folder by type and date, and show me the plan before you move anything."
• "Draft a friendly reply to this email in my voice." (Paste the email.)

Notice the pattern: you say what you want, give it the context (the notes, the folder, the email), and let it work. This is the same skill as writing a good prompt — be clear about the goal, the context, and the format.

Step 4 — Watch it work, and approve the consequential steps

Cowork plans the steps and starts doing them. For anything reversible (drafting, organizing, summarizing), let it run. For anything irreversible or outward-facing — sending an email, deleting files, posting publicly — it should pause and ask. Say yes when you mean yes. This human checkpoint is what keeps Cowork safe: it drafts and prepares freely, but you approve what actually leaves your computer.

Step 5 — Add skills and plugins (optional)

Out of the box, Cowork is a capable generalist. You can make it a specialist by installing Agent Skills — small instruction packs that activate automatically for specific tasks. Anthropic offers free ones, and there are paid profession packs (like the vaults on this site) that turn Cowork into, say, a financial-advisor or bookkeeper assistant. Skills install into your Project and run on the Claude plan you already have — no extra per-use fee.

The habits that make it click

• Start small. Build trust with low-stakes tasks before handing over anything important.
• Give context generously. Cowork is only as good as what you tell it. Paste the real details.
• Keep one Project per area of work. Don't dump everything into one — separate Projects keep the memory and files relevant.
• Review every output that matters. Cowork collapses the busywork; it doesn't take responsibility. For client work, money, or compliance, you read and own the result.

That's the whole on-ramp. Get Pro, make a Project, delegate a small task, approve as you go — and let the scope grow as your trust does.

This article is educational. Claude Cowork outputs should be reviewed by a qualified human before professional use. Claude and Claude Cowork are products of Anthropic; The AI Career Lab is not affiliated with Anthropic.