If you're curious about AI employees, you've probably also found it overwhelming. Every employee seems to need its own skill, its own routine and its own setup, and before you've built one you're looking at a list of twenty.
I didn't start there. I built one ultimate AI employee, a COO. Most COOs wear 20 hats, so it was the easiest and fastest way to get AI doing real work in my business.
Before, I'd sit in one chat window, having one conversation at a time. Now I go through all seven projects, give each task an update, send it to the agent, and come back an hour later to see where it's at. Then I move it to the next phase. I can do that from my phone or my computer.
The real unlock was spending the time building the management system in ClickUp first. Now I allocate multiple tasks to the COO, and the COO picks which skills and which employees under it to use.
What you're building
This is what mine looks like today. Every one of my AI employees sits underneath the COO.
Me
└── COO (runs on Opus)
│ Reads every task I give it & decides: do it, split it,
│ or hand it back to me for a session
│
├── Ops Manager (runs on Sonnet)
│ │ The gritty admin: tidy-ups, lookups, admin drafts
│ │
│ ├── Inbox Assistant
│ │ Sorts my email every hour & drafts the replies
│ ├── Knowledge Base Updater
│ │ Writes the day into my AI brain every night
│ └── Morning Brief
│ Tells me what today is before I'm at my desk
│
├── Blog Manager
│ Writes my blog posts from my briefs
│
├── Scriptwriter
│ Helps me write my Instagram scripts
│
└── ManyChat Manager
Sets up the keyword flows that send people my posts
And this is how I use it, every day.
- I update the task with what I want in the "activity feed" on clickup
- I set the AI EMPLOYEE field to the COO.
- I set the status to "w/ agent to do".
- I walk away.
- When it's done, it's back in my inbox at "w/ samantha for review", with the work at the top of the task and my brief kept underneath.
If I want changes, I comment on the task and set it to "w/ agent for edits". If I'm happy, I close it.
What you need before you start
- The Claude desktop app on a paid plan, with the Claude Code tab.
- ClickUp. Any task manager with custom fields, statuses and comments will work, but the steps say ClickUp.
- Obsidian set up for your AI brain. The COO reads your business from your files every time it picks up a task, so it has to be written down somewhere Claude can read. If you haven't set that up, start with The Co-Founder Setup.
- A computer that's awake with Claude open while you want the COO working.
You don't need to know how to code. Claude writes the one small script this needs, and runs anything that needs running.
Step 1 - Build the workflow first
The workflow is what lets you hand work to the COO and get it back. It's two things in ClickUp: the same statuses on every list, and a field that says which AI employee owns the task. The field says who. The status says whose turn it is.
Open the Claude desktop app, click the Claude Code tab and start a new chat. Click the + at the bottom of the chat and choose your AI brain folder. If you haven't connected ClickUp yet, open Settings, then Connectors, find ClickUp and click Connect. You sign in on ClickUp's own screen.
Then paste this.
I want to set up ClickUp so I can hand work to AI employees, starting with
an AI COO.
Look at my lists & suggest one set of statuses I can use on every list.
It needs to include "w/ agent to do", "w/ agent for edits", "w/ [MY NAME]
for review" & "w/ [MY NAME] for desktop session". Then walk me through
adding those statuses & a dropdown field called AI EMPLOYEE with one
option, COO. One step at a time.
Claude tells you exactly where to click. Statuses and custom fields are set in ClickUp itself, so you'll do the clicking and Claude will tell you what to click.
These are the ones that matter on my lists:
to sort - brain-dumped, not looked at yet
to do - mine, decided
w/ agent to do - an AI employee is working on it
w/ agent for edits - back with the employee, with my notes
w/ samantha for review - done, waiting for me to read it
w/ ss for desktop session - needs me at the Mac with Claude
done
If you want the full version of this setup, with the views and the task brief template, it's in How to Build the Workflow for Your AI Employees.
Step 2 - Write the COO's skill file
The skill file is the COO's job description. It says who it is, what it knows, how it decides what to do with a task, what it hands back and what it never does.
I've turned mine into a template you can download, with none of my business in it: Download the COO skill file template →
Inside the zip is one file, COO-Skill-Template-by-Incuv8or.md. It has every section my COO works from:
- Who I am and What I know: where it finds your business, your clients and how your task manager is organised.
- Triage first: do it, split it for another employee, or hand it back to you as a session, plus a space to list your active employees and how you want tasks triaged.
- Opus or Sonnet: which work stays with the COO and which goes to a cheaper Ops Manager.
- How I decide what the result is, Output format and Hard rules.
- Learned rules: where every correction you give it gets written down, so it doesn't make the same mistake twice.
Unzip it, put the file in your AI brain folder, then in the same chat, paste this.
I've downloaded the COO skill file template & put it in my AI brain folder.
Fill it in with me. Ask me one question at a time for everything in
[square brackets], & use what's already in my AI brain where you can.
Save it as "COO.md" in my AI brain folder & show me before you save.
Claude will ask you things like where your client folders live and which employees you have. If the answer is "none yet", that's fine. The COO does the work itself until you add one.
Step 3 - Set the parameters
The hard rules in the template are the parameters, and they're the part that lets you walk away. Mine can't send anything, pay for anything, delete anything or touch a password. Anything that goes out into the world stops with me.
There are two more rules I'd add before you connect it.
It posts as its own ClickUp user. Mine posts as a separate user called Claude AI Team. When I read a task, I can see what I wrote and what the AI wrote, and so can the employees when they read the task back.
It logs every change. Every time it changes something in ClickUp, it leaves a comment on that task saying what it changed, and adds a line to a list I have called AGENT RUNS. If I ever want to know where and when something happened, it's in one place.
Add two rules to my COO skill. It posts in ClickUp as its own user,
[AI USER NAME], never as me. And every change it makes gets a comment on
that task & a line in my [LOG LIST NAME] list, with a link.
Step 4 - Connect it
Now the COO needs a way to find work without you opening a chat. Mine uses a small Python script called the dispatcher. It checks ClickUp every 10 minutes for tasks with an AI EMPLOYEE set and a "w/ agent" status, and starts Claude on each one with that employee's skill file.
In the same chat, paste this.
I want my COO to pick up work from ClickUp on its own. Write me a small
Python script, the dispatcher, & set it up to run every 10 minutes on
this Mac.
Every run, it checks ClickUp for tasks where AI EMPLOYEE is set & the
status starts with "w/ agent". For each one, it starts Claude with that
employee's skill file & the task. When Claude's done, it comments on the
task with the result & sets the status to "w/ [MY NAME] for review".
My ClickUp key goes in my Mac's Keychain, never in the script & never in
this chat. When it's time, tell me where to create the key & open the
place it goes for me. Explain each step in plain English & do any setup
yourself.
A key is a password for your ClickUp account that a script can use. The Keychain is the password store that's already built into your Mac.
Claude tells you when to do each of these.
- Claude sends you to the page in ClickUp's settings where you create a personal API token. Create one and copy it.
- Claude opens the box where the key goes on your Mac. Paste it there, on your own screen.
- Go back to the chat and tell Claude it's done.
If anything ever asks you to paste a key into the chat, don't. If one does end up in a chat, delete it in ClickUp and make a new one.
What you'll see is Claude telling you the dispatcher is running and when it'll next check. Test it with one small task, like "find my three oldest tasks in to sort and suggest what to do with each". Set AI EMPLOYEE to COO and the status to "w/ agent to do". Within 10 minutes it should start, and when it's finished the task will be at "w/ [your name] for review" with the answer at the top.
Step 5 - How you actually use it
This is the part that changed my days. Update the task, pick the employee, set the status to "w/ agent to do", and walk away. Review it when it comes back at "w/ samantha for review".
The COO reads every task and chooses one of three things. If it's one piece of work, it does it. If it needs more than one employee, it splits it into subtasks and gives each one to the right employee. If it would change something live, it doesn't try. It writes me a brief I can paste into a chat and hands it back.
That third one is what makes me trust it. When my Blog Manager pulled four changes to my morning brief out of one of my voice notes, the COO didn't try to change the live routine in the background. It wrote each one up and handed them back, so I could sit down with Claude and do them with me at the keyboard.
Step 6 - Add a second layer
Once the COO is working, give it someone underneath. My Ops Manager does the gritty admin: ClickUp tidy-ups I describe, lookups, admin drafts, subscriptions and calendar questions.
The COO runs on Opus. The Ops Manager runs on Sonnet, which is faster and lighter, so the admin doesn't use up Opus. The COO keeps strategy, money, client work in my voice and anything ambiguous. The Ops Manager hands anything that needs a judgement call back up to the COO, and anything that would change something live comes back to me as a session. That's the layer of approval, and it means I read and review what's happening before anything live changes.
Now I want an Ops Manager under my COO for the routine admin: tidy-ups,
lookups, admin drafts & simple research. Write its skill file from my COO
skill. It runs on Sonnet. Anything that needs a judgement call goes back
up to the COO, & anything that changes something live comes back to me
as a session. Then add Ops Manager to my AI EMPLOYEE field, & tell the
COO it can delegate to it.
When to use a routine instead
Not everything should go through the COO. Work that has to happen every single day, at the same time, becomes a routine, delegated to its own employee. My morning brief, my inbox assistant and my meeting notes all run that way. The COO is for the work that arrives whenever.
If you're not sure which one a job is, I wrote up the five questions I use: Claude routine or AI employee?
The catch, and the fix
The COO can't do everything in the background. Anything that changes something live, like a routine, a website or your ClickUp setup, needs you at the keyboard. The first time a task comes back as a session instead of done, it feels like it failed.
The fix is to treat that status as part of the system. The COO sets the task to "w/ ss for desktop session" with a ready-to-paste brief, so when I sit down at the Mac I open the task, paste the brief into a new Claude Code chat and it gets built with me there.
The second catch is that the dispatcher runs on your Mac. If the laptop is shut, nothing gets picked up until you open it again. If you want your COO working while you're out, leave the Mac awake with Claude open.
What this looks like as a business
Every founder you support has the problem I had: one chat window and 20 hats. Building them a COO on top of their task manager is a systems job, and you're paid for knowing what to build and in what order. The workflow comes first, then the COO, then the employees underneath.
Start with one
Don't build the dispatcher first, and don't build the Ops Manager first. This week, set up the statuses and the AI EMPLOYEE field, and write the COO skill. Then give it one real task and read what comes back. When that's right, connect it.
If you want to see where the COO fits in the whole team, it's in How I Built an AI Team in Claude.
If you haven't got your AI brain set up, that's the foundation for all of this: start here.
this is the free stuff — imagine what we'd build together
Ready to make your business AI-enabled?
Consulting and one-to-one coaching for founders and freelancers who want AI actually working inside their business — not just talked about.