In October 2025, I couldn't write a function. By April 2026, I had 10 production services running, serving real users daily. No bootcamp. No CS degree. Just a decade of tech curiosity and a refusal to stay on the sideline.
Here's the honest version of how it happened.
I want to be upfront: I didn't learn to code from absolute zero in any meaningful sense. I had 9 years in the tech industry — selling computers, managing an Apple reseller, diagnosing hardware as a Geek Squad field agent. I understood:
What I didn't have: any experience building web applications, any understanding of APIs, any knowledge of JavaScript/Python/Node.js, or any idea how to deploy a service.
First contact with Claude AI. Started using it as a coding tutor, asked it to explain every line. Built tiny scripts — a weather fetcher, a file renamer, a price scraper. Felt impossibly slow.
First real projects. Discovered OpenClaw, an AI-assisted development platform, and started building with Claude directly. Learned about API keys, async/await, error handling. Built the Polymarket trading dashboard — real money at stake, so I had to get the API integration right. Learned Express, REST APIs, ethers.js.
Polymarket dashboard matured. Published my book Tokenomic: REPRICED by AI. The writing process itself was AI-assisted — another proof point that these tools amplify existing knowledge, not replace it.
ESP32 e-ink display project. Jumped from web to embedded (C, ESP-IDF). Completely different world. Built a server-side renderer that pushes weather/stocks/news to a physical e-ink screen. Also built media management tools. Transitioned from OpenClaw to Claude Code — a step change in velocity.
Multi-agent AI platform. This is where something shifted. I drew the system diagram on paper before writing a single line of code. Five boxes, arrows between them, shared filesystem at the bottom. It took ten minutes to draw, and I realized I wasn't looking anything up. The architecture was already in my head. Built my own Telegram bot, then migrated to Claude Code's official channels. Infrastructure maturity: Cloudflare Tunnel, home dashboard, unified toolbox portal, monitoring. 10 services running in production.
I used Claude extensively, but I never copy-pasted without understanding. My rule: if I can't explain what every line does, I don't ship it. This was slower at first, but it meant I actually learned.
The pattern: describe what I want → get code → read every line → ask "why this instead of that?" → modify it → break it → fix it.
I never completed a tutorial or online course. Every project started with a problem I actually had:
It turns out the motivation to fix your own broken thing is ten times stronger than the motivation to complete someone else's exercise.
My first Telegram bot was 50 lines of code with no error handling. It crashed constantly. But it worked well enough to prove the concept, and that momentum carried me forward. I've rewritten every project at least twice.
The skills that transferred surprised me. Nine years of explaining RAM upgrades to confused parents turned out to be the same muscle as writing clear UI text. Debugging a customer's dead MacBook and debugging a crashed Node.js process use the same logic: isolate, reproduce, test.
And running a store — real P&L, real inventory, real staffing — taught me to ask "what does this cost to run?" before "what can this do?" That question saves more engineering time than any framework.
Nobody gave me permission to call myself a developer. I shipped 10 services and then it was just true.
I built everything with AI assistance. I can also explain every line without it. These two facts are not in conflict.
Six months ago, I was fixing printers. Now I'm arguing with my AI agents about code architecture at 2am. I don't know what the next six months look like, but I know the pattern: find a problem, build something, ship it, make it better. That's the whole method. There isn't a secret.
Want to see the system I built? Here's the deep dive on the multi-agent platform.
Gerald Yang — AI Solutions Builder based in Ontario. Currently exploring roles where building things and explaining things are the same job.
Email · LinkedIn · GitHub