Blog · AI and technology
Why I build software with a team of AI agents
I spent twelve years in restaurant operations before I shipped my first app. Here is why I chose to build with AI agents instead of the traditional route, and what that looks like day to day.
In 2007 I started as a crew member at Dunkin'. Over the next twelve years I worked my way up to store manager. I learned how to run a shift, train a team, fix whatever broke at the worst possible moment, and keep customers coming back. I did not learn to code there. I taught myself on my own time, one small project after another.
On October 10, 2026 I started building in public. Since then my studio, SoeSan Holdings, has shipped 28 products. I did not write most of that code by hand. A team of AI agents did, and I want to explain why I chose to work this way when many developers still prefer the traditional route.
The traditional route
The usual path into software is clear: learn a language deeply, then a framework, then the tools around it, and spend years writing code line by line until you are fast. It works, and the people who walk it build real expertise.
It also assumes you have years to spare before your work earns anything. I did not want to wait that long to start building things people could actually use.
What changed
AI agents can now write, test and deploy code. They are not perfect, and they still need someone to point them in the right direction. But the bottleneck has moved. The hard part is no longer typing the code. It is deciding what to build, checking that it works, and making sure it is safe.
Those are the parts I already knew from running a store. A manager does not do every task on the shift. A manager decides what matters, sets the standard, checks the work and takes responsibility when something goes wrong.
How it works day to day
My job is direction and judgment. I decide what we build and why. The agents write the code, test it and put it live. They report back, and I review what they did.
There are a few things I never hand over:
- Anything that involves money
- Anything that touches people's personal data
- Anything published or sent under my name
Those decisions stay with me. Everything else can move quickly, and that is why a single founder can ship at the pace of a small team.
The bottleneck has moved. The hard part is no longer typing the code. It is deciding what to build.
What it has produced
Pennyrow is the studio's flagship. It turns a year of bank and card statements into clean books for tax season, and it runs entirely in the browser, so statements never leave the user's device. There is also Seat Finder for wedding seating charts, Name Stock for the history of every US first name, and more on the products page.
Not every build has been a hit, and that is part of the process. Shipping something real teaches you more than planning ever does.
What comes next
Now I am working on the harder part: turning a fast-moving studio into a real business. That means making products people will pay for, and building websites, web applications and videos for other businesses using the same approach.
I will write here every week about building with AI, what is happening in the industry, and the occasional story from the road that got me here. If you want to follow along, subscribe to the RSS feed or follow me on X.