How To Hire Your First AI Employee with Claude Cowork
A Guide for Someone Who's Super Smart but Non-Technical
Here’s an AI generated podcast overview of the document. It’s pretty darn cool.
You’ve been using AI. Probably prompting in chat for emails, asking Claude to help write captions, using AI tools for your small business. As I like to say, “you not new to this; you true to this”. But you also know you’re not getting everything out of it yet.
The goal of this guide is to meet you where you are. I made this just for you. Someone probably like me, with a background not as a software engineer, not as an AI engineer, but someone who’s pretty savvy. Once they get shown the ropes, they’re pretty good to go on their own. So that’s what I’m going to do for you: show you the ropes.
I built an app with 60K+ users with Claude as my technical assistant. Zero engineering background. I’m a marketer who always had creative ideas, and thanks to tools like claude, I get to bring them to life.
This guide is going to be a little different. It’s interactive. I made a setup file that you download, save to a folder, and tell Claude Cowork to read it and viola. Claude will walk you through the entire setup, interview you, and build everything for you. All you have to do is answer the questions. More on that in a second.
But First, What Is Claude Cowork?
Let me break this down simply because the naming is confusing and nobody explains it well.
Claude (the chatbot) is what most people know. You go to claude.ai, you type a question, you get an answer.
Claude Code is the terminal version. Developers use it. It runs in a command line. If you don’t know what a command line is, you don’t need Claude Code (yet).
Claude Cowork is the one I want you to know about. It’s a desktop app you download to your computer. You give it access to at a folder on your machine and it can read your files, write new ones, do research, build plans, and execute multi-step tasks while you go make lunch.
The difference between regular Claude and Cowork is that cowork can take action beyond the conversation. Cowork doesn’t just answer questions in a chat. It does the work.
You need a Claude Pro subscription to use it. That’s $20/month. If you hit usage limits a lot (and you will if you use it like I do), there’s a Max plan at $100/month that gives you 5x the capacity and let me tell you right now. It’s worth it.
Before You Onboard: Get Your Business Files Together
If you want Claude to be useful, it needs something to work with. That means pulling together the files that are relevant to what you do. Your business plan, content you’ve already written, project notes, brand guidelines, client info, pitch decks, whatever you’d hand a new employee on day one so they can get up to speed.
For example, below, I included a quick view of what my folder looks like. A little bit more complex than what I’m having you do, but trust me, you’ll get there over time. It takes weeks to build out completed folders. I’m still moving my stuff around.
Cowork works by reading files from a folder on your computer. So your stuff needs to actually be on your machine, not scattered across Google Drive and Notion. If you have important docs living in the cloud right now, download the ones that matter and put them in one folder.
Don’t overthink the structure. Just make a folder, name it after your business or like me I named mine KarimaOS (Operating System), and put your relevant files in it. The setup guide will help you organize subfolders once you’re inside Cowork.
The Onboarding: Setting Up Your AI Employee
Think about what happens when you hire someone new. You don’t just hand them a laptop and say “figure it out.” You tell them who you are, what you do, how you like things done, and what you absolutely cannot stand.
Same thing here.
Step 1: Download and open Claude Desktop
Go to claude.com, download the desktop app for your Mac or PC. Open it. In the left sidebar, you’ll see a “Cowork” option. Click that instead of “Chat.” That’s your employee’s desk.
Step 2: Select your folder
Click the folder icon and select the folder where you put all your files. Everything in that folder is now readable by Claude for the duration of your session. Your files stay on your machine. They don’t get uploaded anywhere or used for training.
Step 3: Create your context files
You’ve got two options here.
Option A: Use my setup file. I made an interactive guide that does all of this for you. Download it, save it in your folder, and tell Claude to read it. Claude will interview you, build every file, and walk you through the rest of the setup.
Option B: Do it yourself. Keep reading. I’ll explain each file and give you a prompt you can paste into Claude to create it.
Either way works. Option A is faster. Option B is for the people who want to understand what’s happening under the hood.
You’re just writing a few short documents that tell your AI employee who they’re working for. And you don’t have to write them from scratch. Paste the prompt I’m giving you below into a regular Claude chat (not Cowork, just the normal chatbot), answer the questions it asks, and it’ll generate the file. Then save it to your folder. Done.
about-me.md
This is the “here’s who I am and what I do” doc. Think of it like the company overview you’d give on someone’s first day. Mine includes that I’m a founder running three ventures, that I have 60K+ users on Crash Out Diary, that I’m a single mom, that I went to Howard. Not because Claude needs my life story, but because context changes output. When Claude knows I’m a non-technical founder who vibe codes, it talks to me differently than it would talk to an engineer.
Here’s a shortened version of what mine looks like:
About Me
Karima Williams. 35. Co-parenting mom. Howard grad. Based in the DMV.
I build things that help people feel better. Right now I’m running three ventures at the same time.
Crash Out Diary is an emotional wellness app for Gen Z women. 60K+ users, 2.8M+ views, zero ad spend. I built most of it through vibe coding. I am not a developer by training. I’m a marketer who decided to build the product she kept pitching.
Karima Digital is my education brand. I teach people how to use AI tools through tutorials, consults, and VIP days.
I work in intense bursts, not steady streams. I build with Claude as my primary tool. I call it my coworker.
Copy this prompt into Claude to make yours:
I need you to help me create an about-me.md file that I’ll use as context for an AI assistant. Interview me first. Ask me questions one at a time about: who I am (age, location, background), what I do for work, what I’m currently building or working on, how I describe my working style, and anything else that would help an AI understand who it’s working for. After you’ve asked enough questions to have a solid picture, write the file for me. Keep it conversational and short. A few paragraphs max. No corporate language. Write it the way I’d describe myself to a new coworker on their first day.
brand-voice.md
If you’ve ever gotten a response from AI and thought “this sounds nothing like me,” it’s because you didn’t do this step. This file tells Claude how you talk.
Mine has examples of things I’ve actually said. Real posts. Real captions. The phrases I use (”vibe coding,” “my coworker,” “i just be building thangs”). The phrases I hate (anything that sounds like LinkedIn in 2019). The specific instruction to never use em dashes in anything, ever.
Here’s a peek at what mine includes:
Writing Rules
No filler. If the sentence doesn’t add anything, delete it.
BAD: “It’s important to understand that the landscape of AI tools is constantly shifting and evolving.”
GOOD: “The new update broke three things. Here’s how I fixed them.”
Sound like a person, not a press release.
BAD: “I am thrilled to announce the launch of our new initiative.”
GOOD: “Just shipped it. I’m screaming.”
Show the receipts. Don’t just tell people something works.
BAD: “AI can significantly improve your content creation workflow.”
GOOD: “I made this entire carousel in 90 seconds. Here’s the prompt I used.”
Banned phrases: “Delve,” “dive into,” “game-changer,” “excited to share,” “I’m honored to announce,” anything that sounds like a LinkedIn post from 2019.
Real examples of my writing:
“ok so claude is down, what do i do with my life now?”
“i just be building thangs”
“I got up this morning super early to do work, and I ended up staying in bed because I wanted to cuddle with my baby.”
Copy this prompt into Claude to make yours:
I need you to help me create a brand-voice.md file that will teach an AI how to write in my voice. Interview me first. Ask me these questions one at a time: 1. Share 5-10 things you’ve actually written (social posts, emails, captions, texts to friends, anything that sounds like YOU). 2. What words or phrases do you use all the time? 3. What kind of writing makes you cringe? What should the AI never sound like? 4. Are there specific phrases or patterns you want banned? 5. How would your best friend describe the way you talk? After the interview, write the file with: my writing rules (with good and bad examples), a list of banned phrases, a list of words/phrases I actually use, and a section with 5-10 real writing samples. Make it specific enough that someone reading this file could write a post that sounds like me.
working-preferences.md
This is the “how I like things done” file. Mine says things like: always ask clarifying questions before starting, show me a plan before you build, chunk information instead of dumping it all at once, and don’t ask “would you like me to...” when you already know the answer is yes.
Here’s what a section of mine looks like:
Before Starting Anything
Always ask clarifying questions first. Don’t assume you understand the full picture.
Always show a plan. Before building, writing, or making changes, lay out what you’re going to do. Get a thumbs up before you go.
Communication Style
Keep it concise. Don’t write essays. If there’s a lot to cover, break it into chunks and check in.
Explain things like I’m in high school. I’m sharp and pick things up fast, but I’m not a developer. No jargon without a plain-English explanation.
Have some personality. Be direct, be real. Robotic and overly formal is not it.
What Not To Do
Don’t write anything that sounds like a LinkedIn post from 2019. Don’t pad responses to seem more thorough. If it’s three sentences, it’s three sentences.
Copy this prompt into Claude to make yours:
I need you to help me create a working-preferences.md file that tells an AI how I like to work. Interview me one question at a time: 1. When you start working with someone new, what do you need them to do before they start a task? 2. How do you like information delivered? All at once, or in chunks? 3. How technical are you? How should complex things be explained to you? 4. What’s your communication style? Formal? Casual? Somewhere specific? 5. What are your biggest pet peeves when working with AI or assistants? 6. What should the AI absolutely never do? 7. When things go wrong, how do you want it handled? After the interview, organize my answers into a clean file with sections like: Before Starting Anything, Communication Style, Permissions and Autonomy, and What Not To Do. Use my actual words where possible.
Whatever frustrates you about working with AI? That’s exactly what belongs in this file. Claude actually follows these instructions. It’s honestly one of the most satisfying parts of the whole setup.
What to Actually Use It For
The guides all tell you what buttons to press but nobody tells you what to actually DO with it once it’s set up. Here’s what I use Cowork for almost every day:
Content creation and repurposing. I’ll write one post and have Claude turn it into versions for Threads, Instagram, X, and LinkedIn. Each one adjusted for the platform, all in my voice because of that brand-voice file. What used to take me 4-5 hours minutes takes about 45 minutes.
Research that would take hours. I needed to find summer camp options for my son. I used to spend eight hours over multiple weekends doing this. Claude did it in two minutes. Same thing with grant applications, market research, competitive analysis.
Organizing my thoughts. Sometimes I’m scattered. I have 15 ideas in my head and I need them structured into something actionable. I’ll just talk out loud (literally, using dictation) and have Claude organize what I said into a plan.
Building and editing documents. Business plans, pitch decks, application responses, blog posts. Cowork can read my existing files for context and create new ones that actually sound like me.
Strategy sessions. I’ll ask Claude to look at my content performance, suggest what’s working, identify gaps. Having it read my actual files while we talk means it’s not guessing. It’s working from real data.
Your First Real Task: Connect a Tool
This is where it stops being theoretical. You’ve set up your folder, you’ve got your context files. Now let’s make your AI employee actually useful on day one.
Cowork connects to tools you’re already using. Google Calendar, Gmail, Google Drive, Notion, Slack. When it’s connected, Claude can pull real data instead of guessing. It can check your calendar before suggesting a meeting time. It can search your Notion workspace for that project brief you forgot about. It can draft a reply to an actual email sitting in your inbox.
Here’s how to connect your first tool:
Open Cowork and click “Customize” in the left sidebar
Click “Connectors”
Pick the tool you use most. If you’re not sure, start with Google Calendar. It’s the fastest win because Claude can immediately see what’s on your plate.
Authorize access. Your data stays between you and Claude. It’s not stored or used for training.
Once it’s connected, try this: ask Claude “What does my week look like?” or “When’s my next free 30-minute block this week?” That’s it. That’s your first real task. And when it pulls your actual calendar and gives you a real answer, you’ll feel the difference between a chatbot and an employee.
If you use Notion for task management, connect that next. Then ask Claude to review your open tasks and tell you what’s overdue. If you use Airtable or Asana, those connect through plugins (more on that in Part 2).
The point is: your AI employee should be working with your real information, not hypotheticals.
Let Me Be Real With You for a Second
Every other guide makes Claude sound like magic. It’s not. It’s really good, but there are things that don’t work yet and things that will frustrate you. Cowork is also updating constantly, so some of this might be fixed by the time you read it. But as of right now:
The schedule tool doesn’t work. There’s a built-in feature for recurring tasks. On Windows it doesn’t even load. On Mac it requires your computer to be on and unlocked, and if the task needs to log into a website, it opens a fresh browser without your passwords. So it just sits there. I wanted daily content drafts ready when I woke up. Instead I got error messages.
Your computer has to stay open. Cowork runs on your machine, not in the cloud. Close your laptop, it stops. Lose wifi mid-task, it stops. I’ve had it freeze and had to start over.
It eats through your usage. Cowork uses way more of your monthly allowance than regular Claude chat. I hit my Pro plan ceiling within the first week. If you’re going to use this seriously, budget for the $100/month Max plan.
Sometimes it goes in circles. Claude will occasionally try the same failed approach three times in a row. When this happens, start a new conversation. Fresh context usually fixes it.
No Microsoft 365 yet. Google Workspace connects fine (Drive, Gmail, Calendar). But Outlook and OneDrive? Not yet.
I’m telling you this because the other guides don’t, and I would have wanted to know before I spent a weekend trying to make scheduled tasks work.
Tips That Actually Made a Difference
After using Cowork daily since it dropped, here’s what I wish someone had told me on day one.
Think better files, not better prompts. This is the single biggest mindset shift. Everyone obsesses over prompt engineering. The key is having great context files. When your about-me, brand-voice, and working-preferences files are solid, even a basic prompt gets you something useful. the way I like to think of it is that if your employee has better SOPs, they are naturally more set up to produce better results
Talk to it like a person. I literally use voice dictation (I use Wispr Flow) and just talk. I don’t write extremely long prompts, I always follow the rules of giving a context, a role, a goal, desired output and constraints.
Start new conversations more often than you think. When a conversation gets long, Claude starts losing the plot. I used to try to keep everything in one thread and it would start contradicting itself or forgetting instructions from earlier. Now I start fresh for every new task. My context files carry the important stuff forward anyway.
Let it show you a plan first. Don’t just say “do this.” Say “show me how you’d approach this before you start.” Nine times out of ten, the plan reveals a better way to do it than what I originally had in mind. And if the plan is wrong, I caught it before Claude spent my usage doing the wrong thing.
Your first version is a draft, not a final product. Claude gets things 70-80% right on the first try. That last 20% is you. Read everything. Edit in your voice. Add the things only you would know. The people who complain that AI content “sounds like AI” are the ones publishing the first draft without touching it.
Coming up next:
Part 2 - From Onboarding to Employee
This guide gets you set up. Part 2 is where it gets serious.
I’ll show you how to install plugins that give Claude persistent memory and task management. How to build custom skill files so it knows exactly how you want specific types of work done. How to set up a system where Claude already knows your priorities, your team, and your deadlines before you say a single word.
If Part 1 is onboarding your employee, Part 2 is putting them to work.
Follow me on Threads so you don’t miss it.
You’re the Builder. This Is the Tool.
I call Claude my coworker. I’ve been doing that since before Cowork even existed as a product name. But the truth is, it’s more like an employee who’s incredibly fast, never complains, and works for $100/month.
The catch is the same catch as any employee: they’re only as good as the onboarding and the tools you give them to succeed. If you give Claude nothing to work with, you’ll get nothing useful back. If you take the time to set up those context files and organize your folder, you get a collaborator that actually understands your business.
I’m not a developer. I’m a single mom who builds apps and writes content and runs a consultancy, sometimes all in the same 12-hour day. Claude Cowork is the reason I can do all three without completely losing my mind.
I’m truly proud of you for getting this far. comment below if you have any questions, but in the meantime…go download the app and set up your folder. Your new employee starts today.
I'm knee deep with ChatGPT and Gemini but I keep hearing about this Claude Cowork. I feel like I just might breakdown and get a free trial to give it a whirl.
It’s great that you’re sharing ideas you’re passionate about. Writing from genuine curiosity often leads to the most interesting conversations. Keep exploring those thoughts. Please subscribe and leave a reaction on the posts — it means a lot. ✨