There is a point in almost every small business where the owner becomes the bottleneck.
Not because there isn’t enough software.
Usually there is too much software.
The problem is that someone still has to decide what needs to happen, gather the information, move it between systems, check the work, publish it, follow up, and remember to do it all again tomorrow.
For years, we’ve solved this problem with automation.
If this happens, do that.
Automation is incredibly useful, but I’ve been experimenting with something different at Who Defines You:
AI workers.
Instead of automating individual actions, I wanted to see what would happen if I created specialized digital workers with defined roles, responsibilities, tools, and boundaries.
So I built my own small AI workforce.
The Problem: Too Much Work Depends on One Person
I’ve worked with websites and online businesses for roughly 30 years.
The tools have changed enormously during that time, but one problem hasn’t.
A surprising amount of running a small online business still depends on a human sitting in front of a computer.
Research something.
Write something.
Edit it.
Update the website.
Publish it.
Check whether something worked.
Move information from one system to another.
Remember what needs to happen next.
Each task may only require a few minutes, but together they create a constant stream of small obligations.
Traditional automation can eliminate some of them. But traditional automation works best when the process is predictable.
AI makes another kind of automation possible.
Instead of telling software exactly what button to press, I can give an AI worker a responsibility.
That became the experiment.
The Idea: Build Roles, Not Chatbots
I didn’t want one enormous AI assistant trying to do everything.
That’s essentially the digital equivalent of hiring one employee and making them the researcher, editor, project manager, webmaster, publisher, marketing department, and administrative assistant.
Instead, I started defining roles.
My initial AI workforce included four:
Coordination
Director
The Director coordinates the system. It determines what needs to happen, assigns work to the appropriate worker, keeps track of progress, and escalates decisions that require me.
Discovery
Researcher
The Researcher gathers information, evaluates sources, organizes findings, and prepares useful research for another worker.
Quality
Editor
The Editor reviews material, looks for problems, improves clarity, and helps maintain standards before something is published.
Distribution
Publisher
The Publisher handles approved material and prepares or distributes it through the appropriate publishing systems.
Each worker has a job.
More importantly, each worker has boundaries.
They aren’t four personalities pretending to work together. They’re specialized parts of a larger operating system.
Automation vs. AI Workers
The distinction matters.
Traditional automation generally looks like this:
When X happens, perform Y.
An AI worker can operate more like this:
This is your responsibility. These are the tools available to you. These are the rules you must follow. Complete the work and escalate anything outside your authority.
That opens up a much larger category of work.
A normal automation might copy information from one application to another.
An AI Researcher can be asked to investigate a subject, compare information, determine what’s relevant, organize the findings, and hand the useful material to an Editor.
The exact steps don’t have to be predetermined.
The responsibility does.
The Infrastructure
I built the system around Hermes Agent, an open-source AI agent platform.
Hermes provides the underlying agent capabilities, but Hermes itself isn’t really the interesting part of this project.
The architecture is.
I run Hermes on a small always-on server rather than depending on my primary computer.
That means the AI system can continue operating when my laptop is closed.
The server can run scheduled jobs, communicate with other services, access approved tools, and coordinate work between specialized agents.
Different AI models can also be used for different kinds of work.
A difficult reasoning problem doesn’t necessarily need the same model as a simple organizational task.
That matters because an effective AI system shouldn’t just be intelligent.
It should also be economical.
How Work Moves Through the System
Imagine I want to publish an article.
Instead of personally performing every mechanical part of the process, I can initiate the job.
The Director determines what needs to happen.
Research can be assigned to the Researcher.
The Researcher gathers and organizes the necessary material.
The Editor reviews the resulting content against my standards.
The Publisher prepares approved material for publication.
Christopher → Director → Researcher → Editor → Publisher
But there is an important part of that diagram that shouldn’t disappear.
Christopher.
The purpose of the system isn’t to remove me from my work.
It’s to remove me as the bottleneck.
What I Don’t Want AI Doing
Building AI workers has also forced me to define something equally important:
What shouldn’t be automated?
I don’t want an AI deciding what I believe.
I don’t want it inventing my experiences.
I don’t want it replacing judgment, taste, relationships, or responsibility.
And I don’t want an autonomous system publishing whatever it wants simply because it technically can.
There are places where human approval belongs in the workflow.
Strategy is human.
Taste is human.
Important business decisions are human.
Client relationships are human.
Final approval can remain human.
The AI workers handle the mechanical and cognitive overhead surrounding those decisions.
That distinction is becoming increasingly important as AI systems become more capable.
The question isn’t:
How much can we automate?
A better question is:
What should humans still be spending their time doing?
The Unexpected Benefit: Designing the Business
Something interesting happened while building the workers.
I had to explain how my business actually operates.
You can’t create a useful AI Editor without defining what “good” means.
You can’t create a Publisher without deciding what can be published and where.
You can’t create a Director without determining who has authority to make which decisions.
You can’t automate a messy process without first seeing the mess.
So building AI workers became an exercise in business design.
Every worker forced me to answer:
What is this person’s responsibility?
What information do they need?
What tools should they have?
What decisions can they make?
What requires approval?
What happens when something goes wrong?
Those are useful questions whether the worker is human or artificial.
One Worker Is Often Enough
A business doesn’t need an elaborate AI workforce to benefit from this idea.
In fact, I wouldn’t recommend starting there.
Start with one worker.
Look for an area of the business that is repetitive, clearly understood, and consumes more human attention than it deserves.
Maybe it’s a Research Worker that prepares competitive research every week.
Maybe it’s a Content Worker that turns approved source material into drafts for multiple channels.
Maybe it’s a Website Worker that routinely checks a site for problems.
Maybe it’s an Operations Worker that reviews information from several systems and produces a morning report.
Give that worker one clearly defined responsibility.
Connect only the tools it actually needs.
Establish its boundaries.
Keep important approvals human.
Then watch what happens.
Once that worker is genuinely useful, build the next one.
The Result: More Capacity Without More Complexity
I’m still developing and refining my own AI workforce.
That’s intentional.
I don’t think the goal should be to automate a business as quickly as possible.
The goal should be to remove unnecessary work without creating an even more complicated system to maintain.
The real measure isn’t how many AI agents I can run.
It’s whether I have to spend less of my day managing repetitive work.
If the technology creates another dashboard I have to constantly watch, I’ve probably failed.
If I can give the system a responsibility, walk away, and return when my judgment is actually needed, then it is doing something valuable.
That’s the standard I’m using.
Start with one useful responsibility.
Most small businesses don’t need an AI transformation.
They need one annoying, repetitive part of the business to stop consuming someone’s attention.
That’s where I’d start.
In this experiment at Who Defines You, I explored and built AI workers around the way businesses already operate—from research and content workflows to website operations and internal business processes.
The technology matters.
But the difficult part isn’t installing AI.
It’s deciding what the AI should do, what it shouldn’t do, what tools it should have access to, and where a human still belongs in the process.
That’s business design.
And increasingly, it’s also web design.