AI made building cheap.
It also made it easier to build the wrong thing faster.
Last month, I built my first product with vibe coding.
And I spent more time deciding what not to build than actually building it.
It's an English creative-writing game for kids.
One prompt asks them to rewrite:
"I am tired."
So we can actually see it.
A kid wrote:
"When I'm tired, my eyes get sore, and I shed tears."
The AI teacher didn't rewrite it. Instead, it suggested adding an action — rubbing their eyes or yawning — to make the picture clearer.
AI coach, not ghostwriter.
That's the first hypothesis I wanted to test:
If we break creative writing into small skills — rhyme, rhythm, imagery, metaphor, personification, hook — instead of letting AI write for them, will kids be less afraid of the blank page?
The child writes first. Only then does AI respond.
I think this way because of an argument I had a few years ago.
At a hardware cold-wallet company, our CTO added three new features to a product. The plan was to take it global — at a 50% price increase.
I said no.
"We can build it" and "customers will pay 50% more for it" are two completely different statements.
We went back to what customers actually needed and added features worth paying for.
The product sold out after launch.
Quarter-over-quarter growth: 366%.
That experience taught me:
Viable doesn't mean it runs. It means someone actually wants it.
Today, one person can ship a prototype in days.
That makes product judgment more important, not less.
The question is no longer just:
Can I build this?
It's:
Should I build this? For whom? What assumption am I testing?
That's why I'm learning vibe coding.
Not to become an engineer.
What I love is still building products and business models from 0 → 1.
A business idea used to travel through PM, UI/UX, and engineering before I could test it with a real user.
Now I can turn a hypothesis into something people can actually use — and learn whether I was wrong.
I used to think work had to be 100% before it was fit to publish.
Now I'd rather:
Publish not just the result, but the validation.
The first version is rough. The hypothesis may be wrong.
But never getting real feedback because I was waiting for perfection?
That's the expensive part.
So here it is.
Rap & Poetry Quest
Free. 10–15 minutes to finish.
👉 wordgyms.com/en/
If you have a kid learning English, lend me 15 minutes and let them play on their own.
I want to know two things:
Which level pulled them in? Which level made them want to leave?
No kids? Tell me whether the first level's instructions make sense to you. That's useful too.
In two weeks, I'll share what the drop-off data showed — including the hypotheses it killed.