The most important thing about AI employees isn't the skill, it's the system underneath

Every week there's a new video of someone showing off how many AI agents they've built, or the flashy tool they've wired them all into. It makes it look like the skill is the thing, and like the answer to a business that isn't working yet is one more agent.

I've been saying it all year. The foundation is the most important part. It doesn't matter how many agents you build or which tools you use, because it all comes down to the structure underneath.

Even Anthropic's own guidance points the same way. Their article When to use multi-agent systems (and when not to) says to start with one agent. Their closing advice is: "Start with the simplest approach that works, and add complexity only when evidence supports it."

This is the structure I've used since the beginning, and it's what has made my AI employees so useful. It's how I've been able to scale my business, growing from making $50,000 to $60,000 last year to over $140,000 so far this year.

The foundation, in three parts

The most important thing is not the skill. It's the system underneath, and the foundation of that whole system has three parts.

THE FOUNDATION UNDER MY AI EMPLOYEES

1. THE AI BRAIN
   ├── Expert brain - what I've studied & want my AI to apply
   ├── Knowledge base - me, my business, my clients & where
   │   everything is up to
   └── Organisation - my org chart: the COO, my departments,
       the skill library & the employees that pass the test
       └── Memory & emotional intelligence - where we're at
           right now, & how I work, talk & react

2. THE WORKFLOW MANAGER
   ├── A place to give & receive work - ClickUp
   └── The dispatcher - the small script that keeps the loop
       going, day & night

3. THE CONNECTORS
   └── Access to the tools I work in - Claude connectors
       & API tokens

Skills sit on top of all three.

If your AI brain is thin, your employees guess. If they have no memory, every chat starts from zero. If there's no workflow manager, you're opening a chat and typing every job in yourself. If there are no connectors, Claude can't reach the email, the calendar or the task list where the work actually is. Skills can't fix any of those.

Part 1 - The AI brain

My AI brain is my Obsidian vault, and it has three sections. Every employee reads from it every time it picks up a job, because it starts with no memory of its own.

The expert brain

The expert brain is everything I've studied and want my AI to apply: the frameworks, the courses and the hook library I've built up. None of it is about my business, which is why it sits in its own section. When my Scriptwriter writes a script, it reads the frameworks in here first, so it writes the way I've been taught to write.

The knowledge base

The knowledge base is me and my business: who I am, what I sell, my clients and where every one of them is up to. It's the part that changes every day, so it's the part that goes out of date fastest. There's more on that in the catch below.

The organisation

The organisation is how the work is structured, and it's built with the AI Org Mapper. At the top is my COO, the one AI employee I talk to the most. I brain dump to it, and it decides whether to do the work itself, hand it to an employee or bring it back to me.

Underneath the COO, my business is split into eight departments.

Me (CEO)
└── COO
    ├── BUSINESS OPERATIONS
    ├── CLIENTS
    ├── MARKETING
    ├── SALES
    ├── PRODUCT
    ├── EDUCATION
    ├── TECHNOLOGY
    └── PERSONAL LIFE

My org chart is more complex than most, because I run multiple things. I have an academy that teaches women tech, Tech Queen Academy. I'm building Launchpad, which is my software. And I provide online services, as a COO and in tech and AI enablement. Each of those needs its own home. If you're starting out with one stream of business, you'll need fewer departments than me. Tailor them to your business, which is exactly what the AI Org Mapper does.

Under the departments is one shared skill library. A skill is one process in my business, written down so Claude does it my way, and more than one employee can use the same skill.

I only create an employee when a job passes the four-part test. Anthropic gives three reasons to add another agent: protecting context, running work in parallel and specialisation. Their Claude Code docs on subagents add limiting which tools an agent can use and sending work to a cheaper model. Here's the test I use, built from both.

1. Its own context. A big, focused area of knowledge, like one product.
2. Different permissions or connections. It connects to something the others shouldn't touch.
3. A different model. My Ops Manager runs on Sonnet so the everyday admin doesn't use up Opus.
4. To work at the same time as the others.

If a job doesn't need any of those, it's a skill, held by an employee I already have. The full test is in Do you need another AI employee?

Memory and emotional intelligence

The newest layer in my system sits inside the organisation, and it's the one that's made my agents feel like they actually know me.

Claude doesn't remember anything between chats. Every new chat, and every job the dispatcher starts, begins from nothing. So I gave my organisation a memory, written down in plain documents that every agent reads before it starts.

The memory layer is so every agent knows where we're at at any given time. It has a page that says what I'm focused on this week, what's waiting on me and what each agent is working on. It has a log of what was discussed and decided each day, across every chat. And it has a list of lessons, so when I correct an agent, the correction is written down and every other agent learns it too. A run at 00:45 every night reads the whole day and rewrites the page of where we're at.

The emotional intelligence layer is how my agents get to know me. They learn how I interact, how I talk and how I am, so they can respond to me accurately. During the day, my agents note how I reacted to things, in my own words. Every night, those notes are turned into insights about how I work, each one backed by the notes it came from. Then a one-screen page is rewritten that says how to work with me today, and every chat reads it at the start.

The skills tell an agent how to do the job. The memory tells it where we're at, and the emotional intelligence tells it who it's doing the job for.

Part 2 - The workflow manager

The second part of the foundation is a place to give and receive work. For me, that's ClickUp. Every list has the same statuses, and there's a field on every task that says which AI employee owns it. When I want something done, I set the task to the employee and the status that says it's their turn. When they're finished, it comes back to me for review.

A task manager on its own doesn't make anything happen, though. Something has to notice the task is waiting and start the right employee on it. That's the dispatcher. It's a small Python script that checks ClickUp every 10 minutes and starts the employee the task belongs to, so my AI keeps running throughout the night, not just when I have a chat open.

This is the layer people skip, and it's the one that turns a set of clever chats into a team. Nothing goes live without my approval, and every correction I give becomes a rule the employee keeps.

Part 3 - The connectors

The third part is access. An employee can only work in the tools it can reach, so it needs connections to the tools you live in: your email, your calendar, your task manager and your files.

I do that in two ways. The first is Claude's own connectors. You switch them on in the Claude app and sign in on the tool's own screen, so there's no key to copy anywhere. The second is API tokens, for the tools and jobs a connector doesn't cover. An API token is a password for your account that lets software use it. Mine live in one private place on my Mac, never in a chat and never in a document. My AI employees even have their own ClickUp login with its own token, so when they change something in ClickUp it says it was them and not me.

Keys never go in the chat. If anything ever asks you to paste a password or a key into a message, don't.

Where this sits in the Incuv8or AI Framework

The way I build all of this is the Incuv8or AI Framework. The foundation is the AI brain in steps 1 and 2, and the workflow manager in step 5, with the connectors added wherever an employee needs to reach a tool.

THE INCUV8OR AI FRAMEWORK

1. BUILD YOUR AI BRAIN
   Your business, written down where AI can read it.

2. MAP YOUR ORG
   The AI Org Mapper turns your brain dump into a COO, departments & the
   skills you need, & saves the chart in your brain.

3. BUILD YOUR COO
   The one employee you talk to. It knows your org chart.

4. ADD EMPLOYEES ONLY WHEN THEY PASS THE TEST
   Your COO drafts each one with you: a folder with an agent file &
   skills index. Anything that fails the four-part test is a skill in
   one shared library.

5. HAND OUT THE WORK FROM YOUR TASK MANAGER
   Same statuses on every list, a field for the employee, a dispatcher
   that picks it up.

6. ADD ROUTINES WHERE THEY'RE NEEDED
   Only for jobs that run on a schedule.

If you want the whole framework in one place, it's in You don't need an AI operating system to have an AI team.

What you need before you start

• The Claude desktop app on a paid plan, with the Claude Code tab.
• Obsidian, or any folder on your computer, for your AI brain.
• The AI Org Mapper skill: Download the AI Org Mapper skill →
• ClickUp for your workflow manager. Any task manager with statuses, custom fields and comments will work, but the steps say ClickUp.

You don't need to code to set up the foundation.

Step 1 - Check your AI brain has all three sections

If you haven't set up your AI brain yet, do that first with How to set up your AI brain in Obsidian. If you have, check it's ready to build on.

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. Then paste this.

Read my AI brain. I want to build AI employees on top of it. Check it
has three sections: an expert brain (what I've studied & want applied),
a knowledge base (me, my business, my clients & where everything is up
to) & an organisation section for my org chart.

Tell me which sections are missing or thin, most important gaps first.
Suggest the folders I should create, but don't write anything yet.

What you'll see: a short list of what your employees would have to guess, and the folders to add. Tell Claude which ones to create, then fill in the top gaps before you move on.

Step 2 - Map your organisation

Install the AI Org Mapper: go to Customise, then Skills, then Add Skill, and upload the SKILL.md file from the download. Start a new chat, attach your AI brain folder and type this.

Hey Claude, I want to use /org-mapper to map out my AI org chart.

It asks you what a typical week looks like, what you do on repeat, what you outsource and what has to stay with you, one question at a time. Voice notes are perfect for this. If you run more than one business, tell it, so it gives each one a home.

What you'll see: your COO at the top, your departments underneath, the employees each department needs, the skills each employee uses and the top five to build first, saved in your AI brain as AI ORG CHART.md. The full session is in How to map your AI org chart before you build any AI employees.

Then turn the chart into folders. In the same chat, paste this.

Read my AI ORG CHART.md. Move it into the organisation section of my AI
brain. In that section, create a COO folder, one folder for each
department on my chart & a folder called SKILL LIBRARY.

Don't create any employees or skills yet. Show me the folders when
you're done.

What you'll see: your org chart, a COO folder, your departments and an empty skill library, all inside the organisation section. That's the structure every employee sits in.

Step 3 - Give your organisation a memory

In the same chat, paste this.

In the organisation section of my AI brain, create a folder called
MEMORY with four documents:

- NOW: what I'm focused on this week, what's waiting on me & what each
  agent is working on. Keep it to one screen.
- DAILY LOG: one short entry a day, what was discussed & decided.
- LESSONS: every correction I give, dated, so every agent follows it.
- HOW I WORK: how I like to be spoken to, what helps me & what winds me
  up, in my own words.

Fill in what you already know from my AI brain, then ask me questions,
one at a time, to fill in the rest. Then add a line to every agent file
that says to read NOW, LESSONS & HOW I WORK before starting any job.

What you'll see: a MEMORY folder with four documents, part filled in, and Claude asking you questions about how you work. Answer them the way you'd tell a new assistant, in voice notes if that's easier. Once you have a COO, ask it to update NOW and the daily log at the end of every chat, and when you're ready, turn that into a routine that runs every night.

Step 4 - Set up your workflow manager

Your workflow manager needs three things: the same statuses on every list, a field that says which AI employee owns a task, and a dispatcher that picks the work up. The prompts to set up all three are in How to build the workflow for your AI employees and in steps 4 and 5 of the AI COO post, which is also where you build the COO that the workflow hands work to.

Step 5 - Connect your tools

Start a new chat in the Claude Code tab, attach your AI brain folder and paste this.

Read my AI ORG CHART.md. List every tool my employees will need to work
in, like my email, calendar, task manager & files. For each one, tell me
whether there's a Claude connector for it or whether it needs an API
token.

Walk me through connecting the connectors first, one at a time. For
anything that needs a token, tell me where to create it & open the
private file it goes in for me. I won't paste any keys into this chat.

Claude tells you when to do each of these.

1. In the Claude desktop app, open Settings, then Connectors.
2. Find the tool, for example ClickUp, and click Connect.
3. Sign in on the tool's own screen and approve the access.

What you'll see: a list of your tools, each marked connector or token, and your connectors showing as connected in Settings. Any token goes in the private file Claude opened for you, never in the chat.

Once all three parts are in, you're ready to add employees. Every time you want a new one, ask your COO to run the four-part test first. What goes inside each employee's folder is in How to build a toolkit for your AI employees.

The catch, and the fix

The catch is that the foundation goes out of date. Your business changes, you take on new clients, and your knowledge base still describes last month. Your employees don't know that, so they do the job well for a business that isn't yours any more.

The fix is to keep the knowledge base and the memory current without relying on remembering to do it. Mine has a Knowledge Base Updater that runs every night, reads the day's chats, ClickUp and files, and writes it into my AI brain. You can see how it works in The AI Employee That Updates My AI Brain. After it, the memory run updates where we're at and how to work with me today. Whenever your business changes shape, run the AI Org Mapper again so the organisation section changes with it.

What this looks like as a business

Every business owner you support is looking at the same videos of agents and tools and wondering what they're missing. What they're missing is the foundation: an AI brain that remembers where they're at, a place to give and receive work, and the connections to the tools they already use. Setting that up is the part they can't do for themselves, and you're paid for knowing what to build and in what order.

Start with one

Before you build another agent, open your AI brain and run the Step 1 prompt. Fill in the gaps it finds, then map your organisation and give it a memory.

If you haven't got your AI brain set up, that's the foundation for all of this: start here.