I went almost three years using ChatGPT for almost nothing. Then in July 2025 something flipped. This is the story of the one habit that made all the difference.
One day can change everything
I first heard about consumer AI in late 2022 when ChatGPT dropped. It felt like magic. The first time I asked it something and it answered me…
I had this exact same feeling I had as a kid playing Akinator (that website with the genie who guesses the character you have in your head). I was deep into World of Warcraft back then, so when my mum showed me Akinator I tried to stump it with the most niche character I could think of. It guessed. I was in shock.
Akinanor – blowing minds since 2007
ChatGPT in 2022 hit the same nerve.
Then, funny enough, my interaction with AI was almost zero from 2022 to 2025. I used ChatGPT for the occasional tasks and that was it.
The reason I never went deeper is that when you are given endless choices, making the choice becomes extremely difficult. What you can do with AI sounded like everything at once - and also nothing in particular.
I didn’t know where to start, and I didn’t know why I needed to.
That changed in July 2025. I decided I wanted to build a product with AI. Something simple, end to end, that helps a handful of people. That became my Independence Day from the limits I had put on myself.
That was the beginning of Twitter Screenshot and Soka. I wanted to “vibecode” them but didn’t know even where to start. I had no clue how any of it actually worked. So I decided to just do a first step…
Claude threw twenty new terms at me like React, backend, IDE, Vercel, Supabase, NextJS… in an order that meant nothing to me. I just kept being honest:
Then I went line by line through every term. what I’d heard of, what I hadn’t, what felt scary.
That single reply unlocked everything that came after. And the kind of question that turned out to matter most: not just what I just did, but why each piece exists.
Half the time I couldn’t even formulate the question properly. I’d get a weird answer back, and I’d just ask again.
The hardest part was overcoming myself enough to keep typing “Hey, sorry, I’m stupid and still don’t understand a thing can you please explain why?”
Looking back, I can compress the whole habit into one line:
I always asked why, admitted when I didn’t know, and didn’t rush.
I always asked why. Not just what the AI was suggesting, but why. If the why didn’t land, I pushed back, asked again, asked differently.
I accepted I didn’t know a lot of things. Not as self-deprecation. As a working assumption to learn new things.
I didn’t force myself to be patient about how much time it took to get to know a new concept. The slower I moved, the better prepared I was for the long term. Understanding the concepts beat shipping the app in five minutes.
I literally spent a full weekend dumping every coding word I’d ever heard into a chat. I asked:
“What is all of this? How is it interconnected? Is it all languages, or is it all software? How does it actually fit together?”
It felt slow. It paid off forever.
An Idea For This Week
If you want to try the same thing, here’s what I usually do.
I pick one idea I’ve been sitting on for a while where AI could help. Pick something that excites you: a personal website, a work automation, a side-project, viral agent that everyone talks about. Then it’s simple:
I open Claude/ChatGPT and start a Project (so that context can accumulate). I add one file describing what I want to do and why it excites me. Now the AI has shared context.
I start a chat and ask things like: what are the options? Who are the players? What’s the difference? What would you suggest and why? What are the pros and cons?
When a conversation reaches a natural end, I ask the AI to export a summary as a markdown file. I drop it back into the Project. The next chat picks up from there with full context. One thing I learned the hard way: ten short conversations beat two long ones. Long chats dilute the model.
If your project would require new tools/software – ask about it. It suggested you to use DaVinci Resolve for video editing? Talk to it about how to use it.
Embrace the cringe of silly questions. All the time. It’s the most powerful skill for a human being right now.
This Week I Shipped
I finished a new Chrome extension called Google Meet Transcript. It records private 1-on-1 calls in Google Meet without sending anything to any AI provider. Everything happens locally on your laptop.
Built it for a customer. Then thought it has a great angle for therapists, coaches, and anyone running sensitive 1-on-1s. It’s fully local, it’s discreet (your guest doesn’t see the transcription is on), and it costs me $0 to run.
What’s the one idea you’ve been sitting on that would make you start using AI properly? Reply on this e-mail, I read everything.