A one-person AI company isn't a company with no people in it. It's one person doing the parts only a person can do — choosing the problem, earning trust, making the calls — while AI agents do the repeatable work around them: answering first questions, drafting, scheduling, following up, keeping the books.
That has always been the dream of working for yourself. What's new is how much of the repeatable work an agent can now take on, and how little it costs to start.
Start with a service, not a product#
The quickest way in is to sell something you can already deliver by hand — writing, design, bookkeeping, research, support, a website — and then hand pieces of the delivery to AI. You learn what people will pay for before you build anything, and every hour an agent saves goes straight to you.
A product can come later, once you've done the same job enough times to know which parts never change.
The first three systems#
Build them in this order. Each one pays for the next.
- Intake. An assistant that answers the first questions people ask, collects what you need to know and books a call — with a clear way to reach you, the human, at any point.
- Delivery. Templates, checklists and AI drafting for the repeatable part of the work, so your time goes into the part customers actually pay you for.
- Follow-up. Reminders, invoices, check-ins and the "how did it go?" message that turns one client into three.
Keep the numbers from week one#
Write down, every week: hours per client, what the tools cost, what came in. If it isn't measured, it isn't a business yet — and the numbers tell you which system to automate next.
What stays human#
Choosing who you serve. The first conversation. Anything that needs judgment or trust. And checking what your agents send in your name before customers ever see a mistake.
In public, from here#
Building with AI agents is my day job. The experiments on this site will show the setups and the numbers as they come — what an agent costs, what it saves, and where it breaks.