A year ago, I couldn’t get it built
About a year ago, I tried to build an AI operating partner for small-business owners.
The idea was straightforward: give an owner practical help with the work of running a business. Understanding the finances. Preparing for important conversations. Following up on decisions. Keeping things from falling through the cracks.
I got partway there. But the coding models weren’t capable enough to help me build what I had in mind. Eventually, I put it aside.
I also had a suspicion about what would happen next. By the time the models became good enough, someone else would probably have built it. Or owners would use a general agent platform to assemble whatever help they needed.
Better tools still leave owners with another job
Now the coding tools have improved enough that I can take another serious run at it. And I’m less convinced that the second part happens automatically.
Tools like Cowork, Codex, and OpenClaw open up an enormous range of possibilities. But in the conversations and experiments that have brought me back to this project, I keep encountering the same gap: someone still has to turn those possibilities into useful work for a particular owner.
That’s a bigger job than it sounds.
You have to decide what the agent should do, what information it needs, how it should approach the work, and what a good result looks like. Then you have to check whether it actually helped.
For someone who enjoys building with AI, that can be interesting.
For someone running a business, it’s another thing to manage.
First, you have to know what help you need
Take finances.
Before you can ask a finance agent a useful question, you have to recognize that you need one. Then you have to decide what its job should be.
Should it review your monthly results? Watch cash flow? Look for overdue invoices? Help you understand why revenue is growing but the bank balance isn’t?
An owner might simply be thinking, “We’re busy, but money always feels tight.” Turning that concern into a defined job for an agent takes some understanding of financial management.
A general agent platform leaves a lot of that thinking with the owner. You’re being asked to design the help you need using expertise you may not have.
That’s part of the work I think the product should do for you.
Teach it your business. It should already know its job.
Every business has its own circumstances. A landscaper serving one neighborhood has different needs from a company with crews across an entire metro area.
The owner still needs to explain those differences. But there’s a meaningful difference between teaching someone your business and teaching them how to do their job.
I think business AI should come with more of the second part already figured out. It needs to be a really good consultant.
That’s why I’m building Rex again.
My working hypothesis is that agents with a defined responsibility and a sensible starting method will be more useful to owners than a platform they have to configure themselves.
The owner stays in charge. Rex investigates, does the preparation, and brings back work they can inspect and use.
An experiment with 10–15 owners
I don’t yet know how much will transfer cleanly between businesses, how much tailoring each owner will need, or which jobs will prove useful enough to become a habit.
Those are the things I want to find out by working with actual owners.
I’m looking for a small group of 10–15 business owners, maximum, to help me test Rex. Bring a real piece of work you’d like help with, try what we build, and tell me candidly where it helps and where it falls short.
If you’re interested, leave your details and tell me a little about your business. I’ll follow up personally.