Day 1 of 100: Turning My 4 Companies Into AI-First Companies (While AI Writes This Blog)

I run four companies. Three I own (Scepter Marketing, Ecom Circles and Scepter Commerce) and one I help run (Marapack).

I know. Everybody says grow one business to a million dollars before you start the next one. That's probably good advice. I didn't take it. So here I am.

And here is the honest part. My problem has never been ideas. I have many ideas and I cannot do them all by myself. My problem is execution, and specifically people.

I suck as a manager. I'm bad at training people and making sure they know what they're doing. I'm not even great at finding the right people, though I've been graced over the years with a few who changed everything, and some of them are still with me today. But my whole management style is basically "show up, do your job, and ask me if you get stuck." That works for maybe one person in ten.

So I'm thinking... what if the team member that works that way isn't a person?

What "AI-first company" means to me

Not "we use ChatGPT." Everybody uses ChatGPT.

I mean a company where AI does the actual work and I only get pulled in for the decisions only I can make. In practice that looks like this:

1. Everything that comes into the business (emails, Slack, meetings, my brain dumps) gets turned into a task.
2. Every task gets defined well enough that an AI agent can pick it up cold and finish it without asking anybody anything.
3. Anything that can't be defined yet becomes a question, sent to the one person who can answer it.
4. A heartbeat runs all day on its own, grabs the work that's ready, and hands it to an agent.
5. A different agent checks that work before anything gets called done.

A lot of this goes back to two people I've learned from for years. Ari Meisel at Less Doing taught me to optimize, automate and outsource, and that the better you define a task, the easier it is to hand off. Alex Charfen, formerly of Simple Operations, taught me a lot about running a business with actual structure instead of running it all out of my head. I never managed to get people to run on that structure consistently. It turns out AI will.

The way I think about it now, every task climbs the same ladder: in my head, written down, handed off, runs itself. Most of my work used to stall on the first rung. The whole game is getting it to the last one.

Where I'm at on Day 1

It's off the ground. It's actually shipping meaningful work, and meaningful amounts of it, across different parts of the businesses.

Most of that so far has been operations and admin. Stuff I was still doing myself. It's needed, customers depend on it, but it doesn't always move the needle on revenue. So the next 100 days are about pushing past that:

• Getting sales and marketing running through the same system, not just operations
• Building systems that actually run themselves, not just "AI helps me do it faster"
• Bringing some old projects back to life alongside the core businesses
• Getting partnerships moving, where I plug other people's businesses into this
• Finally learning paid ads. I've always wanted to and I've tinkered here and there, but it's always lost to everything else going on

Yes, AI is writing this blog

This might sound weird. What's the point of a blog if I'm not the one writing it?

That's kind of the point. I haven't been able to keep a blog going in a long time, because it's one more thing on a plate that's already full. If I can get even my own blog off my plate and have it still sound like me, that's the whole idea in miniature. It's written from my words and from what the system actually did that day, and I read every one before it goes up.

Day 1: it looked busy and it wasn't

So here is the thing. Today is the first day the system actually drained work the way it's supposed to, and the reason it hadn't been is a little embarrassing.

The heartbeat was running all day, every day. Over the last week it launched 170 pieces of work. Looks great, right? Except it only looked at 27 of my projects that were on a ranked list. Anything not on that list got skipped. Silently.

When we checked by hand there were 21 pieces of work ready to go across 13 other projects, and it threw away all 21. Zero launches from any of those 13 projects in a week. Meanwhile 8 of the 27 it WAS watching had nothing to do.

The fix was small: go through the ranked list first, then everything else that has work ready. Now it walks 40 projects instead of 27.

The lesson I keep re-learning: a busy dashboard isn't the same thing as work getting done. You have to go look at what actually came out the other end.

Tomorrow: how I actually hand work to AI, and why "hey AI, build me a million dollar business" doesn't work. Soon after, I'll start counting the work the AI finishes on its own and translating it into jobs you'd recognize (VA tasks, admin tasks, developer tasks) so you can see if this is actually moving.

If you want to follow along, follow me on X, where I'll post every day. And if you run a company and you're thinking "I want this in my business," tell me about it here. Curious what you think.

-Matt