Everything you build with AI leaves a trace. Collect enough traces, and you have a path.
The illustration was wrong.
I knew it the moment I opened the file. Color everywhere. Blue background, orange shapes, green lines. Geometric, clean — but not what I wanted. Every illustration already published on the channel was black and white line drawing. A character with orange sunglasses, thick outlines, generous white space, a mr.5pm signature tucked in the corner. What the AI had just produced was the opposite.
I didn't get angry. I paused.
Because in that moment I understood something important. This mistake was mine. I had said "make an illustration." I had never described the existing style. The AI did exactly what it was supposed to do — made a generically good illustration. The problem wasn't the AI's ability. The problem was that I had never written my own visual language down.
That day I did two things.
I placed three existing illustrations side by side and analyzed what they shared. Black and white line drawing. Thick outlines. Orange sunglasses character. Minimal background. Signature bottom right. Then I turned that into sentences — not a formal style guide, just what I understood, put into words. Those sentences went into memory. From the next session on, the AI didn't ask. It just knew.
That was the beginning of the Living Protocol.
I spent more than twenty years in Design Computing research. We had a loop we ran back then. Design something. Analyze it. Extract the hidden pattern. Code it as an algorithm. Reproduce it in the next design. It was manual. Slow. One algorithm could take months to extract.
Make, analyze, extract, record, reproduce.
This was the core of the research. And when I started working with AI, I noticed something strange. The loop was running by itself.
The illustration mistake happened. I analyzed it. Extracted the pattern. Recorded it in memory. The AI reproduced it. What used to take months in the research lab was completed inside a single working session, inside one conversation.
The difference was one thing. This protocol isn't dead.
An algorithm is fixed. Once coded, its form holds. To update it, you analyze from scratch. But a style guide recorded in memory is alive. With each new illustration it becomes more precise. If the character gains more varied expressions, that variation gets recorded. If I decide to change the line weight, that decision is reflected. The protocol follows the work.
That's what "Living" means.
I wrote two color codes into a file. #0D0D0D and #C96442. Black and orange. My brand colors. Now in any session, whichever agent is working, these two colors are the standard. The AI doesn't ask. I don't repeat myself. It just works.
That's what "Protocol" means. Executable, transmissible, reusable.
You are already doing this.
Whether you intend to or not. If you've made the same correction to an AI three times, that's an unextracted pattern. If the AI keeps making the same mistake, that's an unrecorded rule. If you finish a session thinking "next time I should phrase it this way" — that thought is the raw material of a Living Protocol.
The difference is whether you write it down or you don't.
I've protocolized how I write essays too. Scan the source material, shape the structure, forge the actual writing, review it, render it. This essay was made following that process. And things discovered while writing this essay will cycle back into the protocol. It's recursive. Work shapes the way of working, and that way of working shapes the next work.
Traces accumulate into a path.
In my Design Computing research days, we talked about something called "externalization of design knowledge." Taking the tacit knowledge inside a designer's head and making it explicit. The reason it was hard: designers themselves often couldn't articulate how they made decisions.
AI solves this problem differently.
AI watches what you revise. Watches what you choose. Watches the reason when you say "that's not it." Extraction is automated. What once took months of manual work now becomes a single sentence.
But there's a trap here. A protocol has to be recorded to stay alive. Without records, the pattern disappears in the next session. The AI doesn't remember. You forget. The mistake repeats. The correction repeats. The same conversation repeats. A Living Protocol is alive, but without nutrition it dies. The nutrition is documentation.
These days I ask myself one question at the end of every session. "What from today's work can I use again?"
Illustration style. Comment reply approach. Thumbnail composition. Slide layout. Video planning sequence. Essay opening structure. All of these started as a single choice made once. Repeated, they became patterns. Recorded, they became protocols.
My working system now is the sum of these protocols. When I open a new session, the AI reads them. And from the very first output, it already works in my way — the months of trial and error compressed and ready. I don't have to explain from scratch every time. The more protocols accumulate, the faster the next project moves.
What mistake are you repeating?
If you've had the AI revise the same thing three or more times, that's an unextracted Living Protocol. Write that correction into a single sentence. Put it in a file. Have the AI read that file at the start of your next session. Three revisions become one.
That's where to start.
When I saw that illustration mistake, my first reaction was frustration. But that frustration became analysis. The analysis became a record. The record became a protocol. Even now, every time I create a new illustration, that day's mistake is doing its work. Invisibly. Quietly.
Traces become a path. Your mistakes can become a methodology too. But only for the person who writes them down.
What Living Protocol are you building right now?