AI & How I Use It
This site was built together with opencode, an AI assistant that lives in my terminal — and lately with Muse, a personal AI assistant I talk to in chat. This page explains how AI fits into who I am as a learner — and how I keep the understanding mine.
🧠 My Toolbox
opencode — an AI coding assistant in the terminal. It wrote most of this site: the code, the explanations, the walkthroughs of every change.
Muse — a personal AI assistant with its own computer. It drafts posts, designs share images, debugs builds with me — and recently, it can push to this repo with a token I can revoke anytime.
Same rule for both: they write, I review, I decide what stays.
🛠️ Skills I've Built
A small library of custom skills for opencode itself. The first one summarizes YouTube videos by stepping down from cheap to heavy — metadata first, then captions, and local transcription only when neither exists. They live in the cerfly/ai repository, and same rule as the site: it writes, I review and decide what stays.
🤝 How I Use AI
As a patient tutor: I ask questions, it explains, I apply.
As a rubber duck: I explain my problem — to it or out loud — until the fix becomes obvious.
As a pair programmer: it drafts, I review, edit, and own the result.
As an art director: it proposes, I reject, we iterate until it's right.
🚫 Where I Draw the Line
I try not to ship what I can't explain. AI suggests; I decide what stays.
I'll be honest — I don't read every line. I used to review everything before it went live; lately I've let it push directly, with scoped access I can revoke anytime. Either way, the understanding still has to be mine. AI can speed up learning, but it can't do the learning for me.
📋 Want the Details?
More in the posts How I Built This Site — No Frameworks, No Excuses and Three Days With Muse.