Leo Huang
One-person company + an AI team · @kanisleo328
One-person company + an AI team · @kanisleo328
A one-person company. I take client work with AI, build products, and teach people how to use AI properly. The three feed each other: client work shows me real needs, products keep the methods alive, and teaching turns the traps I fell into into shortcuts for someone else.
The question everyone asks first. My income comes from three places: client work, products, and teaching. All three have real paying customers, and all three are still growing.
Client work: brand sites for shops, an internal quoting system for a company, social content for a lecturer. What these jobs have in common is that the client wants one thing solved, and nobody cares who wrote the code. One person plus a set of AI tools, delivering what used to take a small team, measured in days. That is why client work pays.
Products: my own platform and tools (more below). Products are slow, but they keep working while you sleep.
Teaching: 1-on-1, I teach you to turn AI into your team. I don't bill by the hour. I bill by what the result is worth in your hands: after we're done you can do it yourself, I help you get it running, and you can still come back when you get stuck. Details on the services page.
I have one standard for a product: it has to solve one thing for a specific group of people, and that group has to be willing to pay for it. Miss either condition and it's just a portfolio piece.
NaLi Match. Nail and lash salons in Taiwan grew 1.2× in five years, and the market got crowded enough to turn into a price war. Good artists got stuck undercutting each other; customers got stuck not knowing who to book. NaLi does one simple thing: it lets a request and the right artist find each other. The platform now has over five hundred verified artists, with real requests coming in and real services completed every day. It doesn't earn much yet, but every day it produces the fact that someone got helped, and that is how I decide whether a product deserves to continue.
Games. Crispy Kingdom and Immortal Tribulation, two worlds that started from "I want to make my own game", and the reason I started learning to code in the first place. They have players and sponsors. Slow, but always moving.
Open-source tools. Wherever I got stuck making content, I turned the fix into a tool and released it. Take claude-real-video: AI can't actually see the video you hand it, so this tool breaks a video into keyframes and a transcript the AI can really read. MIT licensed, one command to install.
After teaching for a while, I noticed that nine times out of ten the problem isn't "I can't use AI". It's being chased by AI news.
Three ideas, take them:
1. Models keep changing. People who can use them don't panic.
In the first half of this year alone, three big companies released a dozen or so new models, plus one that a government pulled and then let back in.
You can't keep up, and you don't need to. Handing one thing you do every week to AI, and making it smooth, beats knowing a hundred model headlines.
2. AI is the team. You are the boss. It can't read your mind. The clearer you explain, the more complete the material you give it, and the more seriously you check the result, the better the work it hands back.
People who use AI well are really training the skill of being a boss:
break the task down, give context, check the result.
3. Start from one real need. Don't hoard courses.
I've seen too many people buy ten courses, save three hundred tutorials, and then build nothing.
Go the other way: first find one thing you genuinely want solved (mine was "I want to make my own game"), then learn just enough skill to do it.
A real need pulls you through to the end. A course doesn't.
Two years ago I started with a few ChatGPT courses plus Make and n8n, and after that it was a lot of late nights: chatting with AI, asking questions, writing down every "oh, so that's possible".
In July 2025,
because I wanted to build the game world I had in my head, I started coding for real.
The tools kept changing: ChatGPT → Manus → Claude Code,
and I landed on my current setup:
an AI agent that puts every AI to work where it fits.