I’ve spent a lot of nights working on my AI agents lately.
And when I say a lot, I mean the kind of nights where you look at the clock, realize what time it is, and start wondering whether you should even bother going to bed.
I’ve got all these ideas. Businesses I want to build. Tools I want to create. Things I want to automate.
And I’m trying to get everything working together.
But here’s something I’ve been learning.
Just because you’ve got a bunch of AI agents working doesn’t mean you’re getting anything done.
I learned that one the hard way.
Everybody was working. Nothing was working.
A little while ago, I was working with Claude.
I was throwing ideas at him as fast as I could think of them.
Send an agent here. Build this. Fix that. Start another task. Get somebody working on this other thing.
I was excited.
It felt like we were making progress.
Except I wasn’t giving clear enough instructions.
I didn’t have proper standard operating procedures. I didn’t have a good way to make sure every agent had the information it needed. And I hadn’t defined exactly what finished work was supposed to look like.
So the agents were doing things.
Lots of things.
Unfortunately, doing things and doing the right things are not the same thing.
When I started looking at what had actually been completed, things weren’t working the way I thought they were.
I remember sitting on the couch feeling defeated.
I was frustrated. I’d spent all this time trying to build something, and now I was looking at a bunch of work that needed to be fixed.
And I started thinking, man, I don’t even know what I’m doing.
That’s not a particularly enjoyable feeling.
Especially when you’re the one who came up with the whole idea.
But after sitting with it, I started realizing something.
The agents weren’t the whole problem.
A lot of it came down to the way I was managing them.
An AI agent isn’t that different from an employee
Think about what happens when you hire somebody.
You tell them what you need done.
But if you don’t explain how you want it done, what tools they should use, or what a finished job looks like, how are they supposed to know?
Maybe they figure it out.
Maybe they do something completely different from what you had in mind.
Then you get frustrated because they didn’t do what you wanted.
But did you actually tell them?
That’s what I’ve been learning with AI.
You have to give clear instructions.
You need standard operating procedures, or SOPs.
You need to define who’s responsible for what.
And you need a way to make sure those instructions don’t disappear every time you start a new conversation.
Otherwise, you’re not really delegating.
You’re just handing out assignments and hoping everybody reads your mind.
Which would be a wonderful management strategy if mind reading were a standard feature.
Unfortunately, we’re still waiting on that update.
The Army taught me why procedures matter
When I was in the Army, we trained on procedures over and over.
There’s a reason for that.
When things get dangerous, people get scared. They get tired. There’s pressure, confusion, and a lot happening at once.
That’s not when you want to start figuring out what everybody’s supposed to do.
You train beforehand.
You practice the procedures.
You learn your responsibilities.
The goal is to make the right response familiar enough that you can act under pressure.
I’ve been thinking about that a lot while building these systems.
Now, running a business isn’t combat.
But the lesson about preparation carries over.
What happens when your employee is sick?
What happens when the phone won’t stop ringing?
What happens when the owner isn’t available?
Does everyone know what to do?
Or does the whole operation stop until somebody can get you on the phone?
That’s the kind of business I don’t want to build anymore.
A business should be able to operate without you
Here’s how I look at it.
If you own a business, but every decision depends on you, you’ve created a job for yourself.
Maybe it’s a good job.
Maybe it pays really well.
But it’s still a job.
And your boss apparently doesn’t believe in vacation days.
Because your boss is you.
That’s what I’m trying to change.
I want to build businesses that can operate without me having to stand over every little thing.
That doesn’t mean nobody is responsible.
It means the work has a process.
People know what to do. They know where to find the information. They know when to ask for help.
And when something goes wrong, there’s a way to deal with it.
I’m trying to do the same thing with my AI agents.
I want to be able to open one place, give clear instructions, and have the right systems handle the work.
Not spend the entire day moving information between different chats.
Because if I have to manage every little step myself, I haven’t saved any time.
I’ve just hired a bunch of digital employees who need constant supervision.
And somehow I’m still paying for the privilege.
What I’m changing
Right now, I’m working on three things.
First, clearer instructions.
Before I send an agent to work, I need to define what I’m asking for.
Second, better memory.
The agents need access to the correct instructions, decisions, and records. I don’t want to explain the same project from the beginning every time I open a new chat.
Third, a better definition of finished.
If I tell an agent to build something and get it ready to go live, what does that mean?
Does the button work?
Has the workflow been tested?
Are the errors fixed?
Does it work on the phone?
Does it do what the user expects?
And has somebody actually verified it?
Because "the code is finished" and "the product works" can be two very different statements.
I’ve learned that one too.
More than once, unfortunately.
The point is to make those expectations clear before the work starts.
That’s how I want the businesses to operate too.
This is the experiment
I’m calling this The Wild Business Experiment.
I’m testing whether I can use AI, automation, and better business systems to build companies that give me more freedom.
More time with my kids.
More time to travel.
More time to enjoy the things I’ve spent years saying I wanted to do.
And I want to document the process while it’s happening.
Not after everything works.
Right now.
Because some of this stuff isn’t working yet.
I’m still figuring things out.
I’m making mistakes, changing instructions, testing tools, and learning which jobs different AI systems are good at.
Sometimes I spend hours trying to fix something that I thought would take twenty minutes.
That’s part of it.
But here’s what I don’t want.
I don’t want to wake up five years from now with a bunch of businesses, a pile of responsibilities, and no time to enjoy any of it.
I’ve spent enough years chasing the next thing.
I want the business to support the life.
Not consume it.
The lesson I’m taking into this week
If you’re running a business and you’re overwhelmed, look at what depends on you.
Not just what you’re doing.
What are you the only person who knows how to do?
What keeps getting forgotten?
What happens when you’re not there?
Start with one of those problems.
Write down the process.
Make it clear enough that someone else can follow it.
Then test it.
Don’t assume it works just because you wrote the instructions.
And don’t try to automate everything at once.
I’m saying that as somebody who’s been trying to automate just about everything at once.
I can report that it creates some interesting problems.
The goal isn’t to build the most complicated system.
It’s to build something that works.
Something that gives you back a little time.
Because that’s what the business is supposed to do.
Give something back.
I’m still working on it.
And I’m going to keep documenting what happens.
The good stuff, the mistakes, the things that work, and the things that make me sit on the couch wondering what in the hell I was thinking.
That’s The Wild Business Experiment.
We’re building it now.
Later, we’ll have the story.
Go be wild.
—Michael
Leave a Reply