YouMind
Sign in

Complete Guide to Setting Up Claude Cowork: Run Your Business Solo

@nobel_824
JAPANESEJun 29, 2026
279K
232
18
3
578

TL;DR

This guide explains how to configure Claude Cowork as an AI business partner by setting up context-rich folders and global instructions to automate repetitive tasks and content production.

You want to start a side hustle, but time just slips away between your main job tasks. Replying to emails, formatting documents, organizing files. Your hands are moving, but you never get to the work that actually requires thinking. You hear AI is useful, but in the end, you just ask a question in a chat, copy-paste the answer, and call it a day.

Most people just starting a side hustle probably stumble on at least one of these. I was the same way.

Open up Claude, ask a question, copy the response, paste it somewhere else, and close the tab.

It's convenient, but the actual work still remained on your plate.

What changed things for me was starting to use Claude Cowork.

Instead of asking for answers, I started handing over entire "folders" of work. What came back wasn't text, but completed files.

Draft emails, organized folders, reports built from tables. All I had to do was check the final content and give it the OK.

The time I spent on manual work turned directly into time for making decisions.

This article compiles the initial setup steps that side-hustle beginners can copy exactly. Nothing complicated here.

Just create a folder, write about yourself once, and then make a short request.

I'm tatsuki (@nobel_824). I support small and medium-sized businesses in adopting AI, helping them integrate tools like Claude and Codex into their operations, while running Claude myself every day.

The foundation is a post by Rahul (@sairahul1), which has been read over 5.15 million times overseas (original article here). It was excellent, so I've reorganized the steps and content for a Japanese audience.

From "Getting Answers" to "Handing Over Work." That's Claude Cowork.

tatsuki | Claude Code活用支援 - inline image

The Claude most people use is the chat version. You open it, type something, and get a reply back. You paste that text somewhere, and close the tab. This is convenient in its own way, and I use it every day too. However, the manual work still remains with you.

Claude Cowork is a product designed for a different purpose. It runs on desktop apps (Mac / Windows) and the web, and can read, write, and create new files within a folder you specify. It can even create files like documents and spreadsheets, running through multi-step tasks to completion. Anthropic released this as a research preview in January 2026, describing it as a tool that's especially effective for people who aren't developers (Official Claude Cowork).

To put the difference in a nutshell: chat is a conversation, Cowork is more like a colleague. If you ask it to "do some research," chat will tell you how to research, while Cowork will do the research and leave a summary file for you. It's the difference in distance between a conversation partner and a hands-on partner.

Start in 5 Minutes

Start with the installation. The steps are simple.

  • Get the desktop app (Mac or Windows) from claude.com/download
  • A paid plan is required (Pro is $20/month on the pricing page. For heavy use, Max 5x is $100/month, and Max 20x is $200/month)
  • Open the app and select "Cowork"
  • Choose one folder from your computer
  • For complex tasks, switch to the highest-performance model in the settings (at the time of writing, this would be Opus 4.8, etc.). Official: Opus 4.8. Model names get updated regularly, so just remember to pick the one at the top of the list.

Starting with Pro ($20/month) is enough. Cowork works with all paid plans, so there's no need to jump into a more expensive plan right away. You can upgrade when you start using it heavily every day.

The folder you choose here will be the starting point for everything. Cowork reads from this folder, writes to it, and organizes within it.

Create the Folder

tatsuki | Claude Code活用支援 - inline image

Many people open Cowork, type in a request immediately, get back a bland result, and are disappointed. The reason is simple: Cowork doesn't know anything about you yet.

This is something you fix once and never have to fix again. Create one folder on your computer called "Claude Cowork," and set up three sub-folders inside it.

  • ABOUT ME (your information)
  • OUTPUTS (where finished work goes)
  • TEMPLATES (where templates go)

That's it. Next, we'll create the most important part: the content for ABOUT ME.

The 3 Files to Explain Yourself Just Once

The three files you put in ABOUT ME are the heart of this system. They are read at the beginning of every session, so once they're set, you'll never have to explain "who you are" again.

File 1: about-me.md (Your Work and Standards)

What you do, how you think, and what you consider "good" work. You don't have to write this yourself. Ask Cowork to interview you and write down the answers. Open a new session and type this in.

I want to create my "about-me.md".

Ask me questions one by one and interview me. About 15-20 questions.

If my answers are vague, dig deeper and ask again.

Please cover the following points:

  • What kind of work I do and who I interact with
  • How I start a task and what I consider "complete"
  • What "good work" means in my field
  • Things I dislike (phrases, sloppy patterns, AI-sounding expressions)
  • My non-negotiable rules

When the interview is finished, compile everything into one markdown file.

Don't just copy the questions and answers verbatim. Extract only the key points using concise sentences and bullet points.

Keep it to around 2000 characters and save it as "about-me.md" in the ABOUT ME folder.

A .md file is just a regular text file you can open in Notepad. When you ask it to "save it," Cowork will create the file itself. You don't need to save it manually.

The tip here is to answer using your voice instead of typing. Both Mac and Windows have built-in voice input. Speaking allows you to convey much more context than typing. The richer the context, the higher the quality of the files it produces.

Don't make the file too long. I initially overdid my self-introduction, making it nearly 20,000 characters. Cowork spent time reading that every session, which took away from the actual work. When I cut it down to about 2,000 characters, the quality remained the same, but the waste was gone.

File 2: anti-ai-writing-style.md (List of Disliked Writing Styles)

Many people dislike AI-ish writing. I do too. The clichés that pop up everywhere, the balanced symmetrical phrases, the overly long paragraphs. This file lists the things you absolutely don't want Cowork to do when writing.

You don't need to make it perfect from the start. Just add expressions you see in the output and think, "I don't like this," whenever you find them. That alone will gradually make Cowork's writing style more your own.

File 3: my-company.md (Goals and Direction)

This year's goals, the market you're focusing on now, and conversely, things you've decided not to do. Think of the bigger direction, not this week's deadlines. You can interview Cowork for this one too.

I want to create my "my-company.md".

This file is meant to tell Claude what I'm aiming for,

to improve the quality of its daily work decisions.

Ask me questions one by one and interview me with about 6-8 questions.

Please cover the following points:

  • The most important goals for this year (preferably with specific numbers), 2-3 of them
  • Key communication channels or markets right now
  • Things I have currently decided not to do
  • Where I've been spending my time over the last 3 months

When finished, compile everything into one markdown file, aiming for about 1000 characters.

Save it as "my-company.md" in the ABOUT ME folder.

You only update this file when your direction truly changes. Once every three months or so is usually enough.

Use Global Instructions to Have It Read Every Time

Even if you create the folder and the three files, it's useless if you don't teach Cowork how to use them. So, you register a set of instructions that will be executed before every session, just once. Open Cowork's Settings, find the "Global Instructions" field, and write this in there just once.

Before every task, read all files in the ABOUT ME folder:

  • about-me.md: My role, standards, and workflow
  • anti-ai-writing-style.md: My writing rules, words I don't want used
  • my-company.md: My goals, direction, and current focus areas

Read the contents of OUTPUTS and TEMPLATES only when I specify a particular file.

Save finished work inside OUTPUTS, creating a sub-folder for each project name.

When a request is ambiguous, do not fill in the gaps on your own; ask a clarifying question.

Don't over-explain; just finish the task and hand it over.

By entering this once, every subsequent session starts from a state where Cowork "knows everything about you." That's why even a short request of about 10 characters can return results that feel like you wrote them yourself. The context is already loaded in.

The One Simple Prompt Template + The Mechanism to Ask Questions First

tatsuki | Claude Code活用支援 - inline image

You don't need to come up with long instructions every time. Basically, this one template is enough.

I want to do "○○ (what I want to do)". The ideal outcome is "△△".

First, read the ABOUT ME folder.

Then, before starting, ask me clarifying questions.

The last line is subtly effective. Many AI tools, if your instructions are vague, will interpret them on their own, run off, and produce results that look nice but are completely off the mark. When Cowork lacks information, it stops, asks clarifying questions, and often offers choices. You can answer by clicking, so you can convey the direction in about 60 seconds. Instead of struggling to write the perfect instructions in text, it feels like clicking options in a form.

For non-engineers, I think this is where it gets the most comfortable. The pressure of "I have to ask properly" disappears, and it feels closer to giving a verbal brief to a capable assistant.

Hand Over Context with Your Voice

Cowork can read large amounts of material quickly and put together tables instantly. Then it waits for you to finish typing. The speed of speaking is much faster than the speed of typing.

So, for complex requests, I recommend handing them over using your voice. As mentioned earlier, both Mac and Windows have built-in voice input, which is perfectly fine to start with. Just speak the flow of your thoughts, like "Something happened recently, and what I really want to convey is this, but first I need to clarify this premise." Spoken language naturally carries richer context than typed text. If you want more comfort, there are dedicated voice input apps, but they aren't essential.

Habits to Save on Usage (Tokens)

When you use Cowork daily, your plan's usage allowance decreases faster than you might expect. You can use the system smartly here.

A quick terminology note. AI reads and writes text in units called "tokens," and paid plans have limits on the number of tokens. The longer a conversation gets, the heavier each subsequent message becomes because it has to re-read the entire history.

The most effective thing is to not drag out conversations and start new ones frequently. When Cowork gives an unexpected result, instead of chasing it by saying "No, not like that," it's faster and lighter to cut the thread, improve your request, and start fresh. If a session gets long, ask it to summarize just the key points, copy that, and paste it at the start of a new conversation. This keeps the context while keeping things lightweight.

Also, grouping related tasks into a single request reduces the need to re-read the background. Letting cheap models handle light tasks like formatting drafts or spell-checking, while reserving the high-tier models for heavy tasks, has also been effective. And keep the ABOUT ME files small. Since they are read every session, if they become bloated, they consume your usage allowance before you even get to the actual work.

Turn Templates into Assets with the Template Folder

When Cowork creates something you think is "good," turn it into a template right there. Just say this at the end of the session.

Save this as a template in the TEMPLATES folder.

Remove the specific content (actual text), and keep only the structure.

Cowork will strip out the actual content and save only the skeleton: the headings, order, format, and length. Next time you need something similar, you can say "Use the ○○ template from TEMPLATES" and it will create it in the same format.

This folder grows on its own without you having to organize it. Successful structures, effective formats, templates for deliverables that clients praised. They accumulate gradually, becoming assets you can reuse any number of times later.

Add Specialization and Integration with Plugins and Connectors

tatsuki | Claude Code活用支援 - inline image

Plugins are packages that turn Cowork into an expert for specific job roles. They bundle skills, commands, and integration settings. Just install one, and Cowork becomes proficient in tasks for that field. Anthropic released 11 official plugins in January 2026, added 12 more in February, and now has over 20 (covering sales, marketing, customer support, product management, legal, data analysis, etc. See the official repository for the list). To install them, open Plugins from the "+" in the chat window, or go to claude.com/plugins, find one that fits your role, and click "Install." After installing, typing "/" in the chat will show you the plugin's commands.

Connectors are mechanisms that link Cowork with actual tools like Slack, Google Drive, and Notion. Once connected, you don't need to use screenshots or copy-paste; you can reference live data mid-conversation. You can ask things like "Summarize last week's Slack conversations" or "Get the numbers from the estimate in Drive." To connect, open the "Connectors" section in Settings, select the service you want to use, click "Add," and follow the on-screen instructions to log in to that service. Once connected, you can use it every time. Connectors themselves are free on all plans and support 38-50+ services including Slack, Google Workspace, Notion, and GitHub (Official: How to use Connectors). You will need a separate subscription for the service you're connecting to.

You don't need to connect everything from the start. Starting with just the one tool you spend the most time on will make the benefits easiest to feel.

Scheduled Runs to Rotate Your Day

tatsuki | Claude Code活用支援 - inline image

The final piece of the system is "scheduled runs," which automatically execute the same task at a fixed time. For example, every morning at 7 AM, it could classify unread emails by urgency, draft replies for routine ones, and prepare meeting notes from your day's calendar. Having that one summary ready before you head out the door lets you see the day's structure in 5 minutes in the morning. The setup is simple: write down the task you want done and say, "Run this every morning at 7 AM," and it will be scheduled to run automatically at that time. Start with a lower frequency and adjust while looking at the results to feel safe.

What worked well for me was the morning setup and the evening review. Having it read the files I touched that day in the evening and compile "What was done, what's in progress, and what to do tomorrow" into a single sheet provides the briefing material for the next morning. The day loosely connects to the previous one, reducing oversights.

Initially, don't go fully automated. It's better to have it include checks like "First, show me the list" or "Confirm before deleting." Once you're comfortable, you can gradually shift safe tasks to automatic.

Taking It Further: Creating Your Own Personal AI Assistant

This is for those who want to use it more deeply. Plugins aren't just for installing ready-made ones; you can also create your own "specialized assistant" that follows your exact workflow. All you need to do is prepare a single instruction sheet (called a "Skill") that outlines your standard procedure.

For example, for a "Weekly Content Creator," you'd provide a procedure like this.

text
1# Weekly Content Creation Procedure
2
3## Tasks
41. Read about-me.md to recall the voice and target reader
52. Never break the rules in anti-ai-writing-style.md
63. Read the idea memos in the INPUTS folder
74. Fact-check the topic with at least 3 web searches
85. Write a hook in the first sentence that stops the scroll
96. Break the main body into chunks of 100 characters or less
107. Finally, add a closing line aligned with the current goals in my-company.md
11
12## Rules
13- One paragraph is max 3 lines
14- Refer to anti-ai-writing-style.md for forbidden words

The more specific the procedure, the better it works. Instead of "Analyze the data," write it like "Compare last week and this week, calculate the percentage change for each metric, mark ones that moved more than 20%, and write a one-sentence observation." Just transcribe the steps you have in your head. Once you create this, next time you can just say "Write one piece using this week's ideas," and you'll get results with the same quality and voice. Review checkpoints, proposal templates, meeting note formatting. The more you repeat a task weekly, the more worthwhile it is to turn it into a Skill.

And when you combine this with scheduled runs, your whole day starts to run itself. In addition to the morning briefing (from the previous chapter), you could hand over "3 things I want to do this afternoon" at noon, and have it review the day's activities in the evening to prepare the head start for the next day. With these three time blocks connected, the day's flow runs without you having to move your hands. When you reach this point, AI is no longer a convenient chat partner, but an assistant that manages your operations. You don't need to do it all at once. Just add one thing at a time: start with the morning briefing, then the weekly content creator.

Common Pitfalls to Watch Out For

It's a useful tool, but knowing a few things in advance can help avoid mishaps.

  • It's still a research preview. Anthropic itself clearly states "Check before sending." 90% of the time it works fine, but don't send deliverables to clients without reading them first.
  • It consumes usage. Continuing with heavy tasks on Pro ($20/month) might make you feel the limits in about a week. Keep a relaxed attitude like "If it becomes my main work tool, I'll consider upgrading."
  • You are granting read/write permissions. Cowork actually touches the files in the specified folder. So, don't initially choose a broad location like your entire Desktop or Home folder. Operate it within a small, dedicated work folder, and keep sensitive keys and documents outside it. It's safer to start with it running on "require confirmation."

When to Use Chat vs. Cowork, and What It Honestly Can't Do

For simple, one-off questions like "What is the capital of France?", the chat version is faster. To avoid mistakes, think of Cowork as a tool for handing over tasks, not for asking questions. Its strength comes out when you hand over "entire chunks of time-consuming work": creating documents, organizing files, reading multiple sources and summarizing them.

Let me also be honest about what it can't do currently. Cowork doesn't have memory across sessions. It starts fresh every time. This is precisely why we use the ABOUT ME files and Global Instructions to provide context, but for long-running projects, you might feel this gap. Also, it's desktop-based, so closing the app stops any running tasks. These are areas likely to change with future updates.

Summary

Before introducing Cowork, I spent a large part of my day on "manual work." After introducing it, that time shifted towards "decision-making work."

For me, the biggest change was "what came back." Before, I would ask a question, get text, and have to assemble it myself. Now, I give a short brief and get back a document that's nearly ready for the client. Competitive research that used to take half a day now involves just dropping articles into a folder and receiving a summary in the morning.

Before I knew it, the bottleneck shifted from "the amount of manual work" to my own judgment and relationships with people. I think that's what running a company by yourself is all about.

3 Steps to Try Starting Today

  • ☐ Install the desktop app from claude.com/download, create the "Claude Cowork" folder, and the 3 sub-folders: ABOUT ME / OUTPUTS / TEMPLATES
  • ☐ Paste the interview prompt from above, and use voice to answer questions to create your about-me.md
  • ☐ Register the Global Instructions just once, and try asking it to do one task you need this week by saying "Read the ABOUT ME folder and ask clarifying questions before starting."

You Can Skip the Initial Setup by Copy-Pasting

After reading this far, many of you might be thinking, "I understand what to do, but I'm a bit unsure about creating everything from scratch myself." Honestly, the most time-consuming part is this initial setup of the folder and the 3 files.

So, I am offering the contents of this article as a ready-to-copy-paste

"Claude Cowork One-Person Company Starter Kit" for free on LINE.

Here's what's inside.

  • The interview prompts for about-me / my-company, where Cowork asks the questions and fills in the answers
  • The Global Instructions you can just paste directly
  • A collection of request prompts for email sorting, meeting minutes, research, weekly content creation, etc.
  • A SKILL.md template for creating your own personal AI assistant
  • Setup texts for automatically running the morning briefing and evening review
  • A list of forbidden words to remove the AI-ish feel

If you paste them in order from the top, the initial setup will be done in minutes. Cowork will ask questions to fill in any gaps, so it's fine even if you're not good at writing.

▼ Get the "Starter Kit" by registering on LINE

https://t.co/90omRA4UQ7

After joining the Open Chat, send

Starter Kit

from the official LINE account noted in the chat, and you will receive the bonus.

Don't try to do it all alone. Start by getting just one thing working. Take it at your own pace.

Remix in YouMind

Turn one viral article into a full content workflow

Collect the source, decode the pattern, create assets, draft the story, and distribute from one AI workspace.

Explore YouMind
For creators

Turn your Markdown into a clean 𝕏 article

When you publish your own long-form writing, images, tables, and code blocks make 𝕏 formatting painful. YouMind turns a full Markdown draft into a clean, ready-to-post 𝕏 article.

Try Markdown to 𝕏

More patterns to decode

Recent viral articles

Explore more viral articles