Claude Code wrote the code for BeMatrx. I decided what to build, and I tested everything. For eight months I sat down almost every night, described the world I wanted as clearly as I could, tried the result on my own phone, and asked for fixes when something was wrong. The app is now on Google Play and the App Store: ninety cities, careers, a working economy and twelve languages. It cost me about $1,200 and most of my evenings. This is how it happened, including the night everything broke.
Who I am
My name is Sezgin Ç. and I live in Antalya, Türkiye. Until last year I had never opened a terminal, never taken a computer science class, and I thought "repository" had something to do with banking. I'm telling you this so you know that everything below was done without any software training.
There was one thing I was good at, though I never thought of it as a skill: I can explain something clearly enough that it's hard to misunderstand. A lot of working life goes like that. You describe what finished looks like, the work comes back, you point out what's missing, and it goes out again. I didn't know most of software development works the same way. The part I was missing was writing the code.
The idea
The idea came from a question I couldn't stop thinking about: what if the time you spent in an app left something behind? A career that moves forward, a friendship that grows because both of you keep showing up, a city on the other side of the world that slowly becomes yours. When I put my phone down at night, I wanted to have something to show for the evening.
I had no team, no money a startup would call a budget, and no idea how to build any of it myself. What I did have was a very clear picture of what I wanted, and I was slowly realising that this was the hard part.
Meeting Claude Code
A friend mentioned Claude Code in passing. It's Anthropic's coding tool, he said: you talk to it and it writes the code. I figured that was an exaggeration, since most sentences with "AI" in them are. Then one night I opened it, described a welcome screen the way I'd describe it to a designer, and the screen showed up working. Then it explained what it had done and why, in plain sentences.
Two things still surprise me. First, it never once made me feel stupid. I asked what a database is, where data physically lives, and, at two in the morning, whether deleting a file was dangerous. Every time I got a straight answer. Second, and more important: it pushes back. When I asked for something that would have quietly broken the world three weeks later, it told me so and explained why in terms I could follow.
What a night looked like
I never learned to code first and built second. I described, looked, corrected and described again. To test on a real device, I'd prop my phone against a bowl on the kitchen table. Here's an exchange from early on, close to word for word:
Me: When someone opens the city map, it should feel like arriving somewhere, not like a page loading. Give it a short pause before anything appears.
Claude Code: builds it, then explains in plain language what changed and why.
Me: Closer. But on a slow connection that pause will look frozen. Show a small sign of movement right away, and delay only the reveal.
As the months went by, the conversations grew along with the world. Payday came out of something like this:
Me: I want salaries to land on the first of the month, at nine in the morning. Not a number ticking up somewhere in the background. A day you look forward to.
Claude Code: proposes how it could work, then asks two questions I had no answer to: whose nine in the morning, when people live in ninety cities spread across the world's time zones, and what someone hired on the twentieth should get on their first payday.
I hadn't thought of that second question, and it was my product. This happened so often that I stopped being surprised and started relying on it. I'd sit down with an idea, and an hour later it would be in much better shape.
I never read the code it wrote. I looked at results, argued about behaviour, and moved on. People find that the hardest part of this story to believe, but that's how it went.
The night everything broke
There's one night I still think about. An update that had worked fine on my phone an hour earlier wouldn't open on a friend's phone. Where the city should have been, there was a white screen. I didn't know what a stack trace was, and I couldn't even name the problem. So I did what I knew how to do and described the symptom in as much detail as I could: what I tapped, what I expected, what I saw instead, which phone, what time, and what had changed that day.
That was enough. Describing a symptom clearly is a real skill, and it carries over from everyday work to software unchanged. You don't need to know why the oven broke; you need to say what happens when you turn the dial. We found the problem and fixed it. I came away from that night with a rule I followed for the rest of the project: I don't need to understand how the machinery works, but I do need to say very clearly what's wrong and what "fixed" should look like.
The first time it felt real
The moment I really believed in this wasn't a launch day. Someone I'd never met signed up, picked a city they'd never been to, and started building a life there. When I saw the new account that evening, I thought: this isn't only my app anymore. Somebody had moved in.
What it cost
Everyone asks for the number: about $1,200 over the eight months. Most of it went to Claude Code. It stopped feeling expensive by the end of the first week, when I compared it to what a single week of a contract development team would cost. The rest went to servers and a database that grew as the world did, the Apple and Google developer accounts every app pays for, and small items like the domain and email that nobody remembers to count. As of 30 October 2026, the project has more than 9,800 commits, 68 automated checks and 13,590 tests.
It isn't cheap. But a year earlier I'd have told you that what I wanted to build needed a team I could never afford.
What I learned
You don't need to read code to think through a product. You need to know exactly what you want and be able to say why, and that doesn't take a degree. I never knew the right technical term for a bug; describing what happened was enough. Shipping a small, somewhat unfinished version taught me more than waiting for a finished one that never came out would have. The hardest part wasn't the tool, it was saying no. Every feature I turned down made the rest feel more like a place and less like a menu of buttons.
Around month four, I noticed I had stopped saying "the thing I'm trying" and started saying "the world I'm building."
If anyone at Anthropic reads this: somewhere in your logs there's someone asking at two in the morning whether deleting a file is dangerous. Thank you for building a tool that answered him patiently. For me, "I have no software training" turned out to be where things started.
If you're new here
BeMatrx is a living world. Your feed is decided by you and the people you follow, newest first, with no ranking algorithm in between. That was a deliberate decision. I'll keep writing here about what building the app is like.
