How curiosity, crazy ideas, and a 4-hour time box turned me into an expert AI builder
This year I committed to building something every day.
Not launching products. Not finishing apps. Just building - one idea, one session, see what happens.
Sixty ideas later, I can tell you what actually changed: I stopped asking “can I build this?” and started asking “what would happen if I tried?”
The crazy ideas that worked
Early on, I had an idea for an AI-powered radio station. Not a playlist - an actual station. Talk shows with two or three AI personalities discussing topics. Hourly news, sports, and weather. Visuals to go with it.
I didn’t know if it would work. I gave it four hours.
It worked. The personalities are AI-generated agents. The scripts come from AI agent writers. The whole thing runs fully automated, producing continuous content without me touching it.
A few weeks later, I wondered if I could build something like Spore - the game where creatures evolve and compete. I didn’t have game dev experience. I had curiosity and a time box.
Four hours later, I had a working prototype. AI generated all the game components using CSS. The enemies have AI-generated personalities. The in-game messaging is generated. Even the sounds.
Then I turned to something closer to my day job. I’m a UX designer, and design systems are a grind - documentation, tokens, components, keeping everything in sync. I wondered if I could automate the whole thing.
I built a pipeline using Claude Code that generates design systems from scratch. Then I figured out how to create Figma components directly from HTML. The kind of workflow that takes teams months, running in an afternoon.
The numbers
Sixty ideas. Forty-eight became functional. That’s an 80% hit rate.
The key was the time box. Four hours maximum per idea. That constraint did two things: it forced me to learn fast, and it kept me from taking ideas too far before validating them.
Some of those four-hour sessions became real projects. Most didn’t. But every single one taught me something.
The stack
Nothing exotic:
∙Claude Code for most of the building
∙Gemini 2.5 Pro for certain tasks
∙Google and OpenAI speech-to-text for voice features
∙OpenRouter for flexibility across models
The tools matter less than you think. What matters is using them every day until they become fluent.
The real shift
Somewhere around idea thirty, something changed.
I stopped approaching new ideas with skepticism. “That’s too complex.” “I don’t know how to do that.” “That would take weeks.”
I started approaching them with curiosity. “I wonder what would happen.” “Let’s find out.”
That shift - from skepticism to curiosity - is the thing I didn’t expect. I thought I was learning tools. I was actually learning a different way to think about what’s possible.
For designers becoming builders
If you’re a designer wondering whether you can cross into development, here’s what I know now: the gap has closed.
You don’t need to master a framework. You don’t need years of engineering experience. You need curiosity, a time box, and the willingness to try ideas that seem too crazy to work.
Most of them will work.
Originally published on Substack.