I don't know how to write code. Over about six months, I used an AI coding tool called Claude Code to build BeMatrx — a living world with ninety cities, elections, a working economy, and twelve languages — through more than five thousand small changes and about $1,100 in tools and hosting. This is the honest version of how that happened: the working nights, the mistakes, and the moment it stopped feeling like an experiment and started feeling like a place.
Who I am, not a developer
I run a business. I've never opened a terminal for fun, never taken a computer science class, and until last year I thought "repository" was a word for banks. None of that is a humblebrag — it's the actual starting line, and I think it matters that you know it before you read the rest of this, because everything below happened without me learning to code.
What I could always do was describe things precisely. I could tell you exactly how a screen should feel, what should happen when you tap a button, why a number in a wallet mattered more than the number itself. It turns out that's most of what building a product actually is. I just needed something that could turn precise description into working software.
Why I started
I kept opening apps that wanted my attention and gave nothing back. Twenty minutes of scrolling, nothing built, nothing changed, just a faint tiredness. I wanted to build the opposite: a place where the time you spend does something — a career progresses, a friendship deepens, a city you've never visited becomes one you've lived in. Not a feed. A world you actually live the life you choose in.
I didn't have a team, a budget most startups would recognize, or the ability to write a single line of the thing I was imagining. What I had was Claude Code and a very clear picture of what I wanted to exist.
How it actually worked
The process was never "learn to code, then build." It was describe, look, correct, repeat — the same loop you'd use directing a very fast, very literal collaborator who never gets tired of your revisions. A real session looked something like this:
Me: "When someone opens the city map, I want it to feel like arriving somewhere, not loading a page. Give it a beat before anything shows up."
Claude Code: builds the change, explains what it did in plain language, shows me the result.
Me: "Closer. But it pauses too long on slow connections and people will think it's frozen — add a subtle sign of life immediately, delay only the reveal."
Another exchange, months later, closer to how elections ended up working:
Me: "I want a city to be able to elect a mayor. Real votes, a real term, a real title people keep."
Claude Code: proposes a data structure, a voting window, a way to prevent the obvious cheat (voting twice), and asks me three questions I hadn't thought about — what happens to the losing candidates, what happens if nobody runs, how long is a term.
I never read the code it wrote. I read the results, argued with the behavior, and kept going. That's the part people find hardest to believe, and it's the truest part of this story.
The night everything broke
There's a night I still think about. An update that had worked perfectly on my phone an hour earlier suddenly wouldn't open at all on a friend's phone — just a blank white screen where a city should have been. I didn't know what a stack trace was. I didn't know what to even call the problem. All I could do was describe the symptom as precisely as I could: what I tapped, what I expected, what I saw instead, on which phone, at what time.
That turned out to be enough. Not because I got lucky, but because describing symptoms precisely is a skill that transfers from running a business to building software — you don't need to know why the oven is broken to describe exactly what happens when you turn the dial. We found it, fixed it, and I learned the lesson that shaped the next six months: I don't need to understand the machinery. I need to be relentlessly precise about what's wrong and what "fixed" looks like.
The first time it felt real
The moment I actually believed in this wasn't a launch or a milestone. It was smaller than that. Someone I'd never met opened the app, picked a city they'd never been to, and sent their first message to a stranger in it. I watched it happen in real time and felt something I hadn't expected: this wasn't my app anymore. It was a place, and someone else had just moved in.
What it cost — the honest math
People ask for a number, so here's my rough breakdown for the six months, in round figures — not an audited statement, just an honest accounting of where the money went:
- AI coding tool usage (the largest single line): roughly half.
- Cloud hosting and database: a modest, steadily growing monthly bill as the world got bigger.
- App store developer accounts (Apple and Google, both required upfront, both annual): a fixed cost regardless of how big or small the project is.
- Domain, email, and small tools: the smallest line, and the easiest to forget when people ask "what does it actually cost to build something."
All in, it landed at around $1,100. Not free, not nothing — but a fraction of what I'd have guessed a year earlier, back when "building an app" still sounded like something that required a team I couldn't afford.
Five things I learned
1. You don't need to read code to reason about a product. You need to know exactly what you want and be able to say why. That skill was never locked behind a computer science degree.
2. Precision beats vocabulary. I never needed the right technical word for a bug. I needed the right description of what happened.
3. Momentum beats perfection. Shipping the small, slightly-wrong version and fixing the next thing kept me moving. Waiting for the "right" version first would have meant there was still no app.
4. The hard part was never the tool. It was deciding what not to build. Every feature I said no to made the ones I kept feel more like a place and less like a list of buttons.
5. Six months is long enough to stop calling it an experiment. Somewhere around month four, I stopped describing BeMatrx as "the thing I'm trying" and started describing it as "the world I'm building." That shift mattered more than any single feature.
What this is, if you're new here
If this is the first thing of ours you've read: BeMatrx is a living world, not a feed. There's no algorithm deciding what you see — that part was a deliberate choice, not an oversight. You pick a city, a career, a version of yourself, and you live the life you choose. With real people.