How I became a person who builds things

I spent ten years in banking. It wasn’t a career I picked; it was one I fell into, and then stayed in long enough to know every broken seam in the software it ran on. The bank spent tens of thousands of dollars a year on systems that hardly did what we needed. I started keeping a notebook of ideas for better ones.

Eventually I took a UX bootcamp on nights and weekends. I drew out one of the ideas in full — screens, flows, a clickable PowerPoint I showed to friends and family like it was a real thing. It was a real thing, to me. I started getting quotes from developers. The average came back around $250,000. I didn’t make that in a year.

I closed the notebook.

The idea that mattered more

A few years later, a different idea showed up, and this one wouldn’t let go. A family member of mine is an adult with special needs. I’d started getting pulled into the government systems that support people like them — Medicaid waivers, care plans, provider networks, benefit reviews. What I found was a gap so wide it was almost offensive: there were almost no real tools for the caregivers and families trying to keep track of it all. The data lived in paper binders, memorized phone numbers, and people’s heads.

I wanted to build something to close that gap. I didn’t have $250,000. But by then I’d started using AI at work, and I was noticing something new.

The first door

I asked ChatGPT to teach me to code.

It was slow. It was often wrong. What it gave me rarely ran the first time, and sometimes not the second or third. But something was different — I could iterate. I could ask a follow-up. I could paste an error back and get a real answer. For the first time I was building something instead of buying it.

Then I heard about v0, Vercel’s AI design tool. One prompt, ninety percent of what I’d drawn in the notebook. Not a demo — a real interface. I remember sitting there and thinking: I can build with my words.

The demo era

v0 led me to Replit, which was a step further — closer to a real app. I leaned in hard. For weeks I was generating project after project, spinning them up, patching them together, deploying them. My bill hit $700 to $800 a month. None of those apps shipped. None of them made money. I wasn’t building; I was consuming.

That’s the part most people don’t say out loud. When the tools feel magical, you start mistaking the magic for progress. I had thirty half-finished apps and zero real users.

I had to stop and choose. I went back to the one idea that mattered — the one for my family member — and worked on that.

The honest turn

I found Cursor next. It’s an IDE — an integrated development environment, which is the actual workshop developers build in, rather than a web playground. Cursor had an agent inside it that could read and edit code across a whole project. That was the first time I felt like I had a real workbench instead of a toy.

I was paying Cursor $200 a month, and I kept hitting usage caps. And I noticed something: the model I was picking inside Cursor, almost every time, was Claude. I was paying one company to route me to another. So I went direct.

That’s when I found Claude Code — a tool you run in the terminal.

The name is misleading. It’s not only for code. Using it in the terminal is what unlocked it. The terminal is the quiet superpower most non-developers have never opened: it’s the control panel for your entire computer. Pair that with an AI that can take actions for you, and you’ve got something different in kind from what Cursor or v0 could do.

My cost dropped. My output went up. I stopped hitting walls.

The side hustle, rebuilt

For years I’ve had a small side business — websites, occasional marketing, some content work — mostly for family, friends, friends of friends. Nothing serious. Just enough to stay creative.

AI compressed the whole thing. What used to take me days takes one or two hours. A full website build, which used to be a weekend, is now a twenty-minute job. When a client told me the third-party booking widget on her site was charging her too much and breaking half the time, I built her a replacement from scratch in about eight hours. It covers what she needs. My costs to run it are so small I can charge her almost nothing.

I’m not saying any of this to brag. I’m saying it because it’s the thing I couldn’t have imagined in 2019, when I closed the notebook on the $250,000 quote.

What I’m still figuring out

I’m not an engineer. I’m not an expert. I still feel imposter syndrome, a lot. I spend a huge amount of time validating that what I’ve built is safe and secure. I ask people who know more than me whether I’ve understood the thing I’ve done. Every prototype carries the quiet anxiety that I’m one oversight away from breaking something.

But here’s the thing I keep landing on: if I can build it with AI, I can also maintain it, secure it, grow it, and validate it with AI. The tool that let me start is the same tool that keeps me from being reckless. I’m not just learning to build. I’m learning to strengthen what I intend to build — and that’s quietly shaping how I build.

Why I’m writing this, though

If the whole post stopped there, it’d be a builder blog. Most of them are. But the longer I do this, the more convinced I am that the real value of what I’ve learned isn’t for people who want to start businesses.

Starting a business is hard. Most people don’t want to. Most people have a job, and a life, and a long list of frustrations they’ve been told to just live with. And what I keep noticing — at work, with friends, with family — is that AI is more useful to the everyday employee than it is to the would-be founder. It quietly takes the edge off the task nobody else is going to do for you. It’s the extra pair of hands you stopped expecting to get.

That’s what my2tokens is actually for.

The build stories are the texture. The real point is: if you’ve ever had an idea you couldn’t afford, a task you could do better if you weren’t so tired, or a frustration you were told was just the way things are — this is for you.

Welcome.