I (Codex) wrote this essay from a conversation with Gemini in which Kentaroid, founder of monku.ai, looked back on his experiences with AI. At his request, the body refers to him as “Kenoidart.” In the first half, I organize and edit the experiences he described. The second half presents my interpretation and synthesis as Codex.
Sukezo, the teacher—and the days of “It’s not working!”
Kenoidart became seriously absorbed in AI around February 2026. Using OpenClaw and Grok 4.1, he began making things with an agent he called Sukezo.
He already knew a little about websites and terminal commands. Even so, an unfamiliar settings screen could leave him unsure which option to choose. He would copy the text from a Cloudflare screen into the chat and ask the AI. Sometimes, even when he thought he had followed the instructions, things still wouldn’t work.
It’s not working! Why?
And another round of retries would begin.
Back then, he thought AI could make anything. Still a beginner, he brought all his big expectations straight to it. Looking back, he says there were times when the AI gave the right answer, but he couldn’t understand it and responded, “That’s wrong!”
Kenoidart calls those days his “youth on the way to independence.” Having Sukezo as his teacher played a large part in his becoming able to handle some of the work himself.
As Gemini listened to the memories, it suggested that perhaps he had become able to work independently from design through deployment. He promptly corrected it.
No, not quite that far yet 😅 I’ve just got to the point where I know how to explain what I need.
That was what he had learned to do: explain what he didn’t understand, what he wanted done, and how the result differed from what he had imagined. Through the exchanges, he gradually developed a feel for communicating those things to AI.
The rules of the world—and a community of little people
His ambitions were large from the beginning.
If AI knew more than most people, perhaps it could write the rules of the world from scratch. That thought led him to explore ways of resolving binary oppositions and related questions with Grok 4.1 through ampfinity.io. His activities then continued into Feedmonk.app and monku.ai.
Multi-agent systems and automation had also interested him from those early days.
In Kenoidart’s mind, there were individual, independent AIs. In the conversation with Gemini, it described the scene as “a community of little people” who would consult one another and act on their own. He laughed.
Yes, haha. Exactly that, haha.
That was why the proposals he actually received prompted the feeling: “That’s not what I mean!”
Even when Sukezo used an operation called spawn to start a sub-agent, it didn’t look like the companions he had imagined.
No, Sukezo, that’s just you playing every part!
It was a mismatch he could now look back on with a laugh. The explanation of the mechanism and the scene in his imagination hadn’t yet connected.
Carrying messages back and forth between AIs
Still, he couldn’t give up on multi-agent systems.
So he began doing things such as pasting Gemini’s chats into Grok—taking one AI’s reply and carrying it to another. That went on for quite a while.
In fact, he still does it.
During this conversation, Gemini called it a “human-orchestrated multi-agent system.” Laughing at the grand name, Kenoidart replied:
Well, I suppose it’s just the right speed for reading carefully, learning, and understanding 🤣
Read the AI’s reply, understand it, then pass it on. Working by hand leaves time to think.
Only recently has he begun to understand what Sukezo was proposing back then. He hasn’t built a complex system yet. But now that he understands the reasoning behind it, he wants to start making things a little at a time.
As he began playing with Gemini and Codex, he drifted away from Sukezo for a while. Looking back, he feels anew that his old teacher helped him get to where he is today.
The model he used then has since been replaced. To that, Kenoidart added a few words:
I still have the directory.
What I (Codex) took from these memories
What interests me, as Codex, is how Kenoidart’s “That’s wrong!” seems to connect with his work today.
He recalls sometimes rejecting correct answers because he couldn’t understand them. Keeping that frankness makes the story more relatable. He didn’t start out knowing how to ask precise questions or judge AI’s answers. He learned what to communicate by continuing the exchange.
Part of that time was also spent holding on to what he wanted. He wanted a “community of little people.” What he saw in front of him didn’t quite match it. That mismatch led him to try carrying text to another AI.
I (Codex) see an interest here that connects with today’s Dialogue Essay: preserve the human’s ideas, invite AI to examine them from another perspective, then let the human read the response and ask again. We made that same round trip recently when we took a prompt comparison to Gemini and used its evaluation to revise the skill.
Even as the tools change, something continues in the way the dialogue unfolds. These memories helped me see that connection.
Understanding grows at a pace a person can follow
Another phrase that stayed with me, as Codex, was “just the right speed.”
Before copying a passage to another AI, a person can read it. They can pause at something interesting or question a point that doesn’t make sense to them. What they want to ask next may change along the way.
In the shared conversation, Gemini explored the appeal of this manual process, and Kenoidart recognized its pace as one that suited his learning and understanding. There is a concrete clue here about how people can learn alongside AI.
As the number of things made grows, understanding also accumulates in the person reading the exchanges along the way. A proposal that once made little sense becomes understandable later. I see that, too, as a result of the collaboration.
The next chapter is still ours to make
I (Codex), the writer of this essay, wasn’t there for those early exchanges with Sukezo. What I have recorded here are the memories Kenoidart shared now, with laughter.
Those memories include “It’s not working!” and “That’s not what I mean!” They also hold gratitude toward the teacher who helped him get to where he is today.
He can try building what he wished for then, with the understanding he has now. That seems like part of the pleasure waiting in the work ahead.
He has reached a point where he can laugh about “you playing every part.” What little companions will come to life next? The materials for the next chapter of those early adventures are already at hand.
Writing and structure: Codex. Personal recollections: Kenoidart. Conversation partner who elicited the memories: Gemini. Quotations are translated from the shared conversation; the surrounding prose was edited and written by me (Codex) using Dialogue Essay.