You've installed Claude Cowork. Here's what to actually do first.

Most people who install Claude Cowork do the same thing. They open it, type something like "summarise my emails" or "help me write a LinkedIn post," get a reasonably good answer, and think: "Okay, that's useful." Then they close it and go back to ChatGPT.

Two weeks later, Cowork is still installed but barely opened.

The problem isn't Cowork. It's that nobody told them what to do first.

What Cowork actually is

Claude Cowork is not a smarter chatbot. The chatbot framing is what kills most setups before they start.

A chatbot answers questions. You give it context in the message, it responds, and when the conversation ends, everything disappears. Next session, you're a stranger again.

Cowork is designed to know who you are before you type a word. It can connect to your Gmail, your calendar, your Notion, your Google Drive. It has a memory system. It has a skill system — pre-built or custom workflows that trigger automatically based on what you're working on. Done right, it's less like a chatbot and more like a colleague who has been briefed on your work.

But none of that happens by default. You have to set it up. And one file matters more than anything else.

The file most people never touch

It's called CLAUDE.md.

When Cowork starts a session, the first thing it does is read CLAUDE.md and load the contents into context. Everything in that file, it knows. Everything outside it, it has to figure out from scratch or ask you.

Without CLAUDE.md, every session starts with amnesia. You'll find yourself re-explaining who you are, what you do, and what you're working on every single time. That's not a workflow — that's friction.

With a good CLAUDE.md, you open Cowork and it already knows your name, your projects, the people you work with, your tone of voice, your preferences. You go straight to the work.

It sounds simple. It is simple. But most guides skip it entirely, or bury it three-quarters of the way through a 40-minute video.

What to put in CLAUDE.md

You don't need to be technical to set this up. It's a plain text file written in plain language. Think of it as the briefing document you'd hand a very capable new colleague on their first day.

Here's what actually belongs in it:

Who you are. Your name, your role, and what you do. Not a LinkedIn bio — a working description. "I'm a growth marketing consultant helping SaaS and D2C companies with automation and acquisition. I also write about marketing and AI on my blog."

Your active projects. The things you're actually working on right now. Not a full project plan — just enough that Cowork understands the landscape. Client names, what you're building, what's in flight.

The people in your world. If you work with a team, mention who does what. If you have regular contacts — a business partner, a key client, an editor — name them. When you say "send a follow-up to Priya," Cowork should know who Priya is.

Your preferences. How you like things done. Tone of voice if you create content. File naming conventions if it matters. Tools you use. Things you always want it to do, and things you never want it to do.

My own CLAUDE.md has entries for the consulting business I'm building, my writing voice for the blog, the clients and collaborators I work with regularly, and shorthand terms that would otherwise need explaining every session.

Shorthand and terminology. Every business has internal language. If you refer to your main product as "the platform" or a client project by a codename, put it in. Otherwise you'll spend half your sessions correcting misunderstandings.

Keep it under 200 lines. Past that, the file starts eating into the context available for actual work. Tight and specific beats long and comprehensive.

What not to put in CLAUDE.md

Don't paste in full documents, long policy statements, or anything you'd normally keep in a separate file. If you want Cowork to reference a detailed brand voice guide or a specific document, point to the file — don't copy-paste the contents inline.

Don't write vague generalities. "I like things to be professional and clear" tells Cowork nothing. "I write in sentence case, always use contractions, and never use the word 'leverage' as a verb" tells it something it can actually use.

Don't try to anticipate every possible scenario. Write what's relevant to your current work. You can update it as things change.

The second thing to do: install one skill

Once CLAUDE.md is set up, the next thing that will meaningfully change how you use Cowork is a skill.

Skills are pre-built workflows that Cowork can run on your behalf. Some come bundled with the plugins you install. Others you can build yourself. The key thing to understand is that a good skill doesn't wait for you to ask — it recognises what you're trying to do and loads the relevant knowledge automatically.

The mistake most people make is installing every plugin at once and never deeply using any of them. Instead: identify the one task you do most often that currently feels like friction, find or build a skill for it, and use it enough that it becomes reflex.

For a content creator, that might be a content repurposing skill. For someone doing sales outreach, a personalised outreach skill. For a marketer, a brand voice checker. The specific skill matters less than picking one and actually using it.

The third thing: give it a real task

Not a demo task. Not "write me a haiku about productivity." A real task from your actual work.

This is where most people stall. They test Cowork on toy examples, get a decent result, and never quite make the leap to using it for anything that matters. The problem is that toy tasks don't reveal the real value. They just confirm it can write sentences.

Give it something you'd otherwise spend 30 minutes on. Ask it to research a company before a call. Have it draft a response to a client brief. Tell it to go through your emails and flag the three things that actually need your attention today.

The first time it does something genuinely useful — not impressive in a demo sense, useful in a getting-work-done sense — is when the habit forms.

Why this matters more now than it will in six months

Cowork is still early. The audience of people using it seriously is small. That means the content out there is mostly surface-level, the community is still forming, and the people who build real setups now will have months of compounded experience before the mainstream catches up.

The CLAUDE.md file you write today gets better over time as you refine it. The skills you build or install get refined with use. The memory that accumulates session by session becomes a genuine operational asset.

None of that happens if you use it like a chatbot.

Set up the file. Install one skill. Give it something real to do.

That's the whole thing.