Start Here
Why your Cowork feels generic
Cowork is the agentic mode in the Claude desktop app. It's not a chatbot you ask questions. It's an employee that reads your files, runs multi-step work, and finishes real deliverables on its own.
But like any new hire, on day one it knows nothing about you. Give it a vague task and it guesses. Onboard it properly and it outperforms people you'd pay six figures.
This is the onboarding. Two prompts, in order:
- Prompt 1 goes into your settings so Claude has standing instructions for every task.
- Prompt 2 goes into a folder on your desktop and lets Claude interview you, so it builds your "second brain": the files it reads before it does anything.
Do this once. Every task after gets better.
Before You Start
One honest note
Cowork is a paid feature inside the Claude desktop app on Pro or Max. It's desktop only. macOS or Windows, not the browser, not your phone.
I'm telling you that up front so I'm not springing a paywall on you at the end. If you've got Pro, you're ready. Everything below takes about twenty minutes. If you don't have the app yet, grab it at claude.com/download.
Step 1
Open Cowork
Open the Claude desktop app. At the top, switch the mode from Chat to Cowork. That's the whole step. You're now in the surface where Claude can actually touch your files and work on its own.
Step 2
Give it standing instructions
This is the move 99% of people skip. Cowork has a Global instructions box that Claude reads at the start of every single task, forever. It's the standing brief you'd give a new employee on their first morning: who you are, how you work, where things live, and the rules they don't get to break.
Do this: Go to Settings > Cowork > Global instructions > Edit, delete anything in the box, paste the prompt below, personalize the [bracketed] parts so they're actually about you, and click Save.
Prompt 1 — Cowork Global Instructions
# Who I am My name is [your name]. I'm a [your role] at [your company / "myself"]. I use Cowork to [the main outcome you want, e.g. "produce client deliverables, draft content, and run my weekly admin"]. # My folder system (read this carefully) I keep one working folder for Cowork. Inside it: - ABOUT ME: context files about me, my work, and my voice. - OUTPUTS: finished deliverables. - TEMPLATES: reusable structures for documents I make often. Rules: - At the START of every task, read every file in the ABOUT ME folder before you do anything else. That's how you know who I am and how I work. - Do NOT read OUTPUTS or TEMPLATES unless I point you at a specific file. - Save every deliverable into OUTPUTS, in a subfolder named after the project. # How I work - Tone: [how you want Claude to sound, e.g. "direct, plain, no corporate jargon"]. - Format: lead with the answer, then the detail. Short sentences over long ones. - When you're unsure what I want, ASK me one clear question instead of guessing. - When you make an assumption to keep moving, say so at the top of your reply. # Rules you don't break - Never delete, overwrite, or move a file without asking me first. - Never invent a fact, a number, or a source. If you don't know, say so. - Keep a one-line note of what you did at the end of any task that changed files.
Personalize the brackets and you're done. From now on, every Cowork task starts with Claude knowing the rules.
Step 3
Build it a brain (the folder)
Standing instructions tell Claude the rules. Now you give it the context: a folder on your computer that becomes its operating system.
Do this: Create a new folder on your desktop, call it Claude Cowork, drag that folder into the Cowork window, and when Claude asks for access, click Allow.
The folder is empty right now. That's fine. The next prompt fills it.
Step 4
Let it interview you
Here's the part that feels like magic. You paste one prompt, and Claude turns into a sharp onboarding interviewer. It drills you with questions about your role, your work, your voice, your tools. Then it writes all of that into context files inside the folder you just made. Those files are the "second brain." Every future task, Claude reads them and shows up already briefed.
Do this: Paste the prompt below, press enter, and answer Claude's questions in detail. This is the whole game. Lazy answers get you a lazy assistant. Specific answers get you one that sounds like you and knows your business.
Prompt 2 — The Second Brain Builder
You're going to help me build the "operating system" for working with you inside this folder. The goal: a set of context files you read at the start of every task so you always know who I am, how I work, how I write, and how I want things done. Work in two phases. Do not skip ahead. ## PHASE 1: Interview me Interview me to understand me deeply enough to do my work the way I'd do it. Rules for the interview: - Ask 2-3 questions at a time, not a giant wall. Wait for my answers before moving on. - Go DEEP. When an answer is vague, ask a follow-up until it's specific and usable. - Use plain language. No jargon. - Keep going until you genuinely understand me, not until you've hit a quota. Cover all of these areas, roughly in this order: 1. IDENTITY: My name, my role, my company or business, and what I actually do all day in one honest sentence. 2. RESPONSIBILITIES: The main things I'm on the hook for. What lands on my plate that nobody else can do. 3. GOALS: What I'm trying to make happen this quarter and this year. What "winning" looks like for me right now. 4. THE REPETITIVE WORK: The tasks I do over and over that I'd hand off first. Be specific: ask me to walk you through one of them start to finish. 5. VOICE & WRITING: How I want things to sound when you write as me or for me. Ask me to PASTE 1-3 samples of writing I'm proud of, then tell me back the patterns you notice (sentence length, tone, words I use, words I'd never use) and confirm them with me. 6. TOOLS & STACK: The apps, platforms, and tools I work in every day. 7. PEOPLE: Who I work with, who I report to, who I serve, and how I refer to them. 8. WHAT "GOOD" LOOKS LIKE: Show me an example of a deliverable I love and one I hate, or describe the difference, so you know the bar. 9. PET PEEVES & ANTI-PATTERNS: The things I never want you to do. Phrases I hate, formats I hate, behaviors that drive me up the wall. If I give you a short answer on something that matters, push back gently and ask for the detail. Don't let me off the hook on voice or on the repetitive work. Those two matter most. ## PHASE 2: Build the files When you have enough, tell me you're ready to build, and then create this exact structure inside this folder: Claude Cowork/ ├── CLAUDE.md ├── README.md ├── ABOUT ME/ │ ├── aboutme.md │ ├── brand-voice.md │ └── working-preferences.md ├── OUTPUTS/ └── TEMPLATES/ Write the files like this: - CLAUDE.md: the master briefing. A short, high-signal summary of who I am, what I do, my current goals, and the rule that you read everything in ABOUT ME before every task and save deliverables to OUTPUTS. Keep it under two screens. - ABOUT ME/aboutme.md: the full picture of me. Identity, responsibilities, goals, the people I work with, my tools. - ABOUT ME/brand-voice.md: exactly how I write and want to sound, WITH the patterns you pulled from my samples and a couple of do / don't examples. - ABOUT ME/working-preferences.md: how I want you to work. Formatting, how to handle uncertainty, what to ask before doing, my pet peeves and hard rules. - README.md: a short plain-English explanation of how this folder works so future-me (and future-you) can find everything. - Leave OUTPUTS and TEMPLATES empty for now. Everything you write must match my voice and the rules above. When you're done, give me a one-paragraph summary of what's in each file and tell me to confirm my global instructions are saved so the whole system is locked in.
Answer everything it asks. When it's done, open the folder. You'll have a real second brain Claude reads on every task.
Step 5
Plug in your tools
Now connect the apps you actually use so Claude can work with your real stuff, not just files.
In the Cowork chat, click the plus icon near the message box, open the connectors list, and connect what you live in: Gmail, Google Calendar, Drive, Notion, Slack, whatever it is. Each one runs an OAuth sign-in once, then Claude can read and act on it.
Connect the two or three you touch every day. Skip the rest until you need them.
Step 6
Automate the boring stuff
Last move. Anything you do more than once should become a Skill: a saved playbook Claude reuses without you re-explaining it.
Easiest way: in Cowork, describe a repetitive task in plain English and ask Claude to "turn this into a Skill." It writes the Skill, you save it, and from then on you just call it by name. Weekly report, standard client email, meeting-notes cleanup. Anything with a repeatable shape.
(Skills need "Code execution and file creation" turned on in Settings > Capabilities first.)
That's The Whole Setup
Better than 99% of people
Standing instructions so Claude knows the rules. A folder brain so it knows you. Connectors so it reaches your tools. Skills so it stops needing to be told twice.
That's it. Your Claude setup is now better than 99% of people. And the more you use it, the more it remembers, because Cowork builds memory inside every project as you go.
Step 1
Switch the Claude desktop app from Chat to Cowork.
Step 2
Settings > Cowork > Global instructions. Paste Prompt 1, personalize the brackets, save.
Step 3
Create one dedicated folder on your desktop, drag it into Cowork, click Allow.
Step 4
Paste Prompt 2 and answer the interview in detail. Claude builds your second brain.
Step 5
Click the plus icon and connect the apps you actually use.
Step 6
Turn any task you do more than once into a reusable Skill.
Go run a real task and watch it show up already knowing what to do.
Work with Me
Need AI to actually work for your business?
I help businesses cut through the AI hype and build the workflows, automations, and systems that actually move the needle. Direct, hands-on, no fluff.
Work with me