I’ve got a pretty simple idea.
I want to build businesses that make money without needing me to sit in front of a computer all day.
I’d like to spend more time with my kids. Travel. Get outside. Go kayaking. Maybe enjoy some of the things I’ve spent years working toward.
You know, unreasonable demands like having a life.
So I’ve been experimenting with AI.
Not just asking ChatGPT to write an email or make a to-do list.
I’m trying to figure out how to use different AI tools, automated systems, and clear business processes to run real companies.
And I’m finding out that there’s a pretty big difference between what AI can do in a demonstration and what happens when you ask it to run part of your business.
The demonstration is usually impressive.
Then you try to use it on a Tuesday when three things are going wrong.
That’s when things get interesting.
Experiment One: Can AI handle the interruptions?
When I ran my roofing company, there was always something pulling me away from what I was doing.
I’d be meeting with a homeowner, working with an adjuster, or dealing with a job.
Then the phone would ring.
Maybe it was another customer.
Maybe somebody wanted an inspection.
Maybe it was a question that could have been answered in thirty seconds.
But I still had to stop what I was doing.
And if I couldn’t answer, I’d have to remember to call back.
I’ve already told you how well my old system of writing phone numbers on pieces of paper worked.
Let’s just say the paper industry probably appreciated my support.
So one of the biggest things I’m testing is an assistant that can handle some of those interruptions.
Here’s what I want it to do.
When a customer calls, it should be able to answer common questions about the business.
It should know our services, where we work, and how appointments are handled.
If someone wants to schedule a call or inspection, it should help them do that using real availability.
If a customer sends an email or text, it should help route the message or prepare an appropriate response.
Eventually, I want it to handle things like payment requests or deposits when the business has approved that process.
But here’s the important part.
I’m not saying all of that is working perfectly today.
That’s what we’re testing.
And I don’t want an assistant making promises to customers that the business can’t keep.
It shouldn’t invent prices, book appointments that don’t exist, or tell somebody their insurance claim is approved.
It needs to know what it’s allowed to do.
And when it doesn’t know, it needs to bring the question to a person.
The goal isn’t to get rid of customer service.
It’s to make sure customers get help while the owner is busy doing the work.
Experiment Two: Can different AIs work together?
Here’s something else I’ve been learning.
Not every AI is equally good at every job.
I’ve been using ChatGPT, Claude, and other tools for different parts of my businesses.
Some are better at certain kinds of planning.
Some are useful for technical work.
Some help with research.
And some are better at writing in the way I actually want to sound.
I found that out while working on articles for Wild Built Studio.
I’d been trying to get the writing right with one assistant.
We went through several attempts.
It wasn’t terrible writing.
But it didn’t sound like me.
It sounded like somebody explaining marketing from a conference stage.
And I’ve spent enough time around marketing to know that nobody needs another article written like that.
So I tried a different approach.
I worked with another assistant, gave it my actual stories, explained how I talk, and started editing with it.
That worked better for the kind of writing I wanted.
Now we’re using those conversations to help design PATCH, a writing assistant that learns how someone speaks and writes.
The lesson wasn’t that one AI is always better than another.
It was that I needed to match the tool to the job.
Just like hiring people.
You wouldn’t hire somebody to run your books just because they’re good at selling roofs.
Well, you could.
But your accountant might have some questions.
I’m trying to learn which systems handle which jobs best, then give each one clear responsibilities.
That’s the experiment.
Experiment Three: Can we stop losing information?
This might be the most important one.
I’ve had conversations where we made decisions, worked through problems, and figured out what needed to happen next.
Then I’d open a new chat.
And suddenly I was explaining everything again.
What we’re building.
What we already decided.
What wasn’t supposed to change.
What had been tested.
What still needed work.
At some point, I realized I was spending too much time reminding the tools how to do the jobs I’d already assigned them.
That’s not much of an improvement over managing employees who forget their training every Monday.
So we’re working on a shared knowledge system.
I want the important instructions, decisions, and procedures saved somewhere the right tools can access.
That includes the original source information, not just a short summary that leaves out half the details.
We’re looking at ways to use a knowledge vault and clear standard operating procedures to keep everything organized.
An SOP is just a written process for doing a job.
It tells someone what to do, how to do it, and what a finished result should look like.
And that’s something I learned to value long before AI.
In the Army, we trained on procedures so people knew their responsibilities under pressure.
In business, the same basic idea applies.
If the owner isn’t there, the work still needs to get done.
If somebody new joins the team, they need to know how the company operates.
And if an AI agent starts a new task, it needs the right instructions.
I’m trying to make those things repeatable.
Not because I enjoy writing instructions.
Believe me, there are more exciting ways to spend an evening.
But because I don’t want to explain the same thing fifty times.
Experiment Four: Can one place run the whole operation?
This is where I’m trying to bring everything together.
Right now, business owners use a lot of different tools.
Email.
Calendar.
CRM.
Website.
Booking systems.
Payment systems.
Project management.
And now AI assistants.
Each one might work fine by itself.
But if you have to spend your entire day moving information between them, you haven’t really solved the problem.
You’ve just built yourself a more expensive collection of tabs.
I want one place where I can see what matters.
What needs my attention today?
What work is being done?
What’s waiting for approval?
Which customer needs a response?
What’s finished?
What’s stuck?
And what should happen next?
The goal is to spend less time coordinating the work and more time making decisions that actually need me.
I’m also thinking about how this could help other business owners.
A contractor shouldn’t have to become a software expert just to keep track of appointments and customers.
A creator shouldn’t need five different systems to publish an article.
And somebody running several businesses shouldn’t have to explain the same instructions to every assistant.
That’s what we’re trying to improve.
But there’s a catch.
Putting everything in one place doesn’t automatically make it simple.
You can build one giant system that’s harder to use than all the smaller ones.
That’s why we’re testing the experience as we go.
If I can’t use it without getting frustrated, I probably shouldn’t be asking somebody else to pay for it.
A revolutionary concept, I know.
How do we know whether any of this works?
This is the part I’m trying to get better at.
It’s easy to get excited when something looks like it’s working.
An agent finishes a task.
A workflow runs.
A new screen looks good.
Everybody celebrates.
Then you try to use it, and the button doesn’t work.
Or the information is wrong.
Or the assistant forgot something important.
So I’m learning to define what success looks like before we start.
For the customer assistant, that means checking whether it gives accurate answers, handles appointments correctly, and knows when a person needs to step in.
For the writing assistant, it means checking whether the article sounds like the author, uses real information, and doesn’t repeat something already published.
For the knowledge system, it means opening a new conversation and seeing whether the right instructions and decisions are still available.
And for the overall business, it means something even more important.
Am I getting time back?
Am I spending less time fixing mistakes?
Can the work move forward without me watching every step?
Those are the results I care about.
We don’t have all those answers yet.
But at least we’re getting clearer about what to measure.
Why I’m showing you the experiment
I could wait until everything works.
Then write a story about how I built these businesses using AI and made it all look easy.
That would probably make a cleaner story.
It also wouldn’t be the whole truth.
Right now, I’m making mistakes.
I’m learning how to give better instructions.
I’m finding out which tools work well together and which ones don’t.
Some nights I make progress.
Other nights I sit there wondering how something that looked so simple became a six-hour project.
I want to document both.
Because I think other business owners are going to run into the same problems.
And maybe they can learn something from the things I’m figuring out.
You don’t have to know everything before you start.
But you do have to pay attention to what happens after you start.
If something isn’t working, find out why.
Change the process.
Test it again.
And keep the parts that actually help.
That’s what I’m doing.
I’m building the systems while I’m building the businesses.
And I’m trying to make sure the businesses eventually give me the time to enjoy the life I’m building them for.
That’s the whole point of The Wild Business Experiment.
We’re practicing right now.
We’re documenting what happens.
And we’ll see where it takes us.
Go be wild.
—Michael
Leave a Reply