Ray Fu, ex-Meta senior engineer and AI automation educator

Ray Fu

I'm an Ex Meta Senior Engineer that makes content and teaches OpenClaw and AI Automations.

stan.store/raycfu

How to Run a One-Person Company with Claude Cowork

(Full Setup Guide)

Don't want to figure this out alone? I walk members through every step inside the community. Join the Skool → skool.com/raycfu

Most people open Claude, type a prompt, get a generic answer, and close the tab. Next time they start from scratch. They re-explain who they are, what they do, how they like things written, and what they are working on. Every single conversation. That is not an assistant. That is a freelancer with amnesia.

The fix is 3 files and one folder. You set it up once and Claude knows who you are, how you work, and what you need before you say a word. From that point forward every session starts with full context about your life and your business. You stop being the person who manages Claude and start being the person who makes decisions while Claude handles the production.

This guide walks you through the exact folder structure, the exact files, the one prompt that runs everything, and how to automate your full day.

WHAT YOU NEED

Claude Desktop app downloaded from claude.com/download. Mac and Windows both supported. You need a paid plan at $20 a month minimum.

A voice dictation tool like Wispr Flow. This is optional but highly recommended. When Claude interviews you for the about-me file your spoken answers come out way more honest and detailed than typed ones. Wispr Flow is free and turns your voice into text in real time.

STEP 1: CREATE THE FOLDER STRUCTURE

Create a folder on your computer called Claude Cowork. Inside it create 3 subfolders:

About Me: This is where your 3 identity files live. Claude reads this folder at the start of every session automatically.

Outputs: This is where Claude saves everything it creates for you. Drafts, reports, emails, plans, whatever the task produces.

Templates: This is where you save reusable templates for things you create regularly. Email formats, report structures, proposal outlines. Claude references these when creating new work so the output matches your standards.

Open Claude Desktop and go to the Cowork tab. Point Cowork at your Claude Cowork folder. Now Claude has access to everything inside it and reads the About Me folder automatically at the start of every session.

STEP 2: BUILD THE 3 FILES THAT CHANGE EVERYTHING

These 3 files are the entire reason Claude stops giving generic output. Each one teaches Claude something different about who you are and how you work.

FILE 1: about-me.md

This file tells Claude who you are as a person and a professional. Instead of writing it yourself which usually produces a stiff corporate bio, let Claude interview you. The spoken answers give 3 times more useful context than anything you would type.

Open Cowork and paste this prompt:

"I want you to interview me to build a comprehensive about-me file. Ask me 20 questions one at a time. Cover: what I do for work, who I work with, what kind of output I create most often, what I care about in my work, what frustrates me, what good work looks like to me, what bad work looks like, how I communicate, what my goals are right now, and anything else that would help you understand me well enough to produce work I would be proud of.

Wait for my answer before asking the next question. Push back if my answers are vague. Ask for examples when I give generic responses.

After all 20 questions, compile everything into a single about-me.md file under 2,000 tokens. Only keep the information that would actually change how you write, think, or make decisions for me. Cut anything generic. Cut anything that sounds like a LinkedIn bio. Keep only the lines that pass this test: if this line disappeared would the AI work differently?

Save the file to my About Me folder."

Turn on Wispr Flow or whatever voice dictation tool you use and talk your answers instead of typing. The interview takes about 15 to 20 minutes. By the end Claude has a tight file that captures how you think, what you value, and how you operate.

FILE 2: anti-ai-writing-style.md

This file tells Claude everything it is never allowed to do when writing for you. This is the file that kills the AI smell in everything Claude produces.

Create this file manually in your About Me folder. Start with this template and add to it every time you see something in Claude's output that makes you cringe:

"NEVER USE THESE WORDS OR PHRASES: Leverage (as a verb) Utilize Streamline Robust Comprehensive Cutting-edge Game-changing Revolutionary Delve Furthermore In conclusion It is worth noting At the end of the day In today's world Moving forward Circle back Synergy

NEVER DO THESE THINGS: Do not start emails with 'I hope this email finds you well' Do not start responses with 'Great question' or 'Absolutely' or 'Certainly' Do not use em dashes Do not use more than one exclamation mark in any piece of writing Do not use bullet points unless I specifically ask for them Do not write paragraphs longer than 3 sentences Do not hedge with phrases like 'it depends' or 'there are many factors' without then being specific Do not summarize what I just said back to me before answering Do not add a conclusion paragraph that just repeats the introduction

FORMAT RULES: Short paragraphs. 3 sentences max. Use numbers instead of vague quantities. Write '3 hours' not 'significant time.' Contractions are fine and encouraged. Write 'don't' not 'do not' in client-facing content. Active voice always. Write 'we shipped the feature' not 'the feature was shipped.'"

This file starts small and grows over time. Every time Claude produces something that feels off, open this file and add the specific thing that bothered you. After a few weeks the file is deeply personalized and Claude's output sounds nothing like AI anymore.

FILE 3: my-company.md

This file tells Claude about your business, your goals, and what you are focused on right now. This is the file you update most frequently because your priorities change.

Create this file in your About Me folder:

"COMPANY OVERVIEW: Name: [your company or project name] What we do: [one sentence description] Who we serve: [your customers or audience] How we make money: [your revenue model] Current size: [team size, revenue range, whatever is relevant]

CURRENT PRIORITIES (update weekly):

  1. [Most important thing this week]
  2. [Second most important]
  3. [Third most important]

ACTIVE PROJECTS: [Project name]: [one sentence status] | Next action: [specific next step] [Project name]: [one sentence status] | Next action: [specific next step] [Project name]: [one sentence status] | Next action: [specific next step]

KEY RELATIONSHIPS: [Name]: [role, how you work together, what they care about] [Name]: [role, how you work together, what they care about]

QUARTERLY GOALS:

  1. [Goal with specific measurable target]
  2. [Goal with specific measurable target]
  3. [Goal with specific measurable target]

THINGS CLAUDE SHOULD KNOW: [Any context that affects how Claude should approach work for you. Industry terminology. Internal processes. Things that are sensitive. Preferences about how you communicate with specific people.]"

Update the Current Priorities section every Monday morning. That takes 5 minutes. Everything Claude does for the rest of the week reflects your actual priorities because it reads this file before every session.

STEP 3: THE ONE PROMPT THAT RUNS EVERYTHING

Stop writing long complicated prompts trying to get Claude to understand what you want. You only need one prompt format from now on:

"I want to [describe the task] so that [what success looks like]. Read my About Me folder first. Then use AskUserQuestion to clarify before you start."

That last part is the feature that changes everything and almost nobody knows it exists. Instead of Claude guessing what you meant when your prompt is vague, it stops and generates a clickable form with specific options you can tap through in about 60 seconds. The AI is prompting you now instead of you wrestling with a text box trying to write the perfect instruction.

You click through the form. Cowork builds a plan based on your answers. You approve the plan. And it creates the actual deliverable in your Outputs folder.

Examples:

"I want to write a cold outreach email to a potential client so that they respond and book a call. Read my About Me folder first. Then use AskUserQuestion to clarify before you start."

"I want to create a weekly report for my team so that everyone knows what shipped and what is blocked. Read my About Me folder first. Then use AskUserQuestion to clarify before you start."

"I want to draft a proposal for a new client project so that they understand the scope, timeline, and investment. Read my About Me folder first. Then use AskUserQuestion to clarify before you start."

The formula never changes. What you want plus what success looks like plus read about me plus clarify before starting. That is it.

STEP 4: AUTOMATE YOUR FULL DAY WITH 3 SCHEDULED SESSIONS

This is where Claude goes from an assistant you talk to into a system that runs your day without you initiating anything.

MORNING SESSION (7:00 AM):

Set up a scheduled task in Cowork by typing /schedule and pasting this:

"Every weekday at 7:00am run this workflow:

  1. Check my email for everything that came in since 5pm yesterday. Categorize each email into three buckets: action required, FYI only, or can ignore.
  2. For anything that needs action, draft a suggested response matching my communication style from the about-me file. Keep responses under 100 words unless the topic needs more.
  3. Check my Google Calendar for today's meetings. For each meeting note who is attending and any preparation I should do based on recent emails or documents related to that meeting.
  4. Check my current priorities from my-company.md and tell me the single most important thing to focus on today.
  5. Save everything to a file called morning-brief-[today's date].md in my Outputs folder.

Keep the entire briefing under 500 words. I want to read it in 3 minutes with my coffee."

You wake up every morning and there is a complete briefing on your desktop. Emails sorted. Responses drafted. Calendar prepped. Priority identified. You did nothing.

AFTERNOON SESSION (12:00 PM):

This is the one session you initiate manually. At noon you open Cowork and brief it on your afternoon tasks using the one-prompt formula:

"I want to [afternoon task] so that [what success looks like]. Read my About Me folder first. Then use AskUserQuestion to clarify before you start."

Claude executes while you handle the meetings, the client calls, and the decisions that only you can make. The production work runs in parallel with your decision-making work instead of competing with it.

END OF DAY SESSION (5:00 PM):

Set up another scheduled task:

"Every weekday at 5:00pm run this workflow:

  1. Review what I worked on today based on any files created or modified in my Outputs folder.
  2. Check my email for anything that came in since this morning's briefing that still needs a response.
  3. List what got completed today, what is still pending, and what carries over to tomorrow.
  4. Update my current priorities in my-company.md if anything shifted based on today's work.
  5. Save everything to a file called end-of-day-[today's date].md in my Outputs folder.

Keep it under 300 words."

That end-of-day document automatically becomes context for tomorrow morning's briefing. So every day builds on the previous day. Continuous loop with zero gaps. Nothing falls through the cracks because the system tracks everything whether you remembered to or not.

HOW TO MAKE IT BETTER OVER TIME

Week 1: Run the system as described. Notice what Claude gets right and what feels off.

Week 2: Every time Claude produces something that does not sound like you, add a new rule to anti-ai-writing-style.md. Every time Claude misunderstands your priorities, update my-company.md with more specific context.

Week 3: Save your best outputs to the Templates folder. The next time you ask Claude to write something similar tell it to reference the template. Output quality jumps immediately because it has a concrete example of what good looks like.

Month 2: You have a deeply personalized system. The about-me file captures who you are. The anti-ai file has 30 plus specific rules. The company file reflects your real priorities. The templates folder has examples of your best work. Claude's output at this point is dramatically different from generic AI because it has been calibrated against your actual preferences over 4 weeks of real use.

Don't want to figure this out alone? I walk members through every step inside the community. Join the Skool → skool.com/raycfu

Explore more free guides