A factory of one: building a software studio solo with AI agents

If you looked in my projects folder you would think a small team worked here. A multi-tenant content platform. A tool for bands to run their operations. A financial decision layer for solo founders. A photographer’s business suite. A retro band site with an actual merch store. On and on.

There is no team. There is me, at night, and a method. The studio those products sit under is called Modus Future, and the one-line pitch I use with myself is blunt: my output, in a bucket.

Here is how one person runs a factory.

The method: waves and briefs#

I do not build a project in one long push. I break it into numbered waves, and each wave into briefs. A brief is a tight, self-contained slice of work with a clear definition of done, small enough to hand off and review in one sitting.

Then I hand each brief to an AI agent (these days that lives in Claude Code, run from inside my editor) working in its own branch. I review what comes back the way I would review a teammate’s pull request: does it do the thing, is it safe, would I be happy to own this in six months. Good briefs merge. Bad ones get sent back.

It is less “AI writes my app” and more “I run a team of one and the team is tireless.” The skill is not typing. It is decomposition, review, and taste.

Why the consistency is the point#

The quiet advantage is that the same discipline runs across every product. Multi-tenant from day one. A dedicated security-hardening wave, every time, because the defender in me will not ship a login page to the internet on a hope. The same handful of stack choices so I am never relearning my own conventions.

That sameness is what lets one person hold twenty things in their head. The portfolio is wide, but it rhymes.

The honest catch#

Here is the part I have had to be truthful with myself about: building is the fun trap. Starting a new product is a dopamine hit. Waves fly by, the code piles up, everything reaches “feature-complete.”

And feature-complete is not launched. It is not a single real user. The factory is extremely good at producing; the frontier, the genuinely hard part, is shipping one of these to people and letting the numbers tell me the truth. Breadth is easy to manufacture. Depth, and revenue, are earned one product at a time.

A factory that never opens the doors is just a very organised hobby.

So the current chapter is not building more. It is picking the strongest one, pointing everything at launching it, and being data-driven about what actually lands rather than what was fun to make.

Where the AI line sits, for me#

People ask where I draw the line on AI-written code. Honestly, I have not found one. I treat it as a real collaborator, not a fancy autocomplete with rules bolted on. I keep the decisions I can check, the architecture, the data model, anything security-critical, and I happily let it do the parts I could do but would rather not.

Backend is where I keep my hands. The front end, I suspect, is getting quietly commoditised by these tools anyway. The data, the systems, the awkward architecture questions, those keep their weight, and those are the parts I enjoy.

One studio, many futures. A factory of one, learning in public which door to open first.