正名 · Zhèngmíng
Twenty-five hundred years ago, asked what he would do first if handed a government, Confucius gave an answer that sounded pedantic and was not: 必也正名乎 — "it would surely be to rectify the names." Fix what things are called, and the rest becomes possible. Leave the names wrong, and no amount of effort downstream will hold, because effort is spent describing a world that isn't there.
I run a stack of AI agents — more than two dozen, each a named specialist: a career strategist, a financial architect, a veterinarian for my dog, an advocate who speaks for me in rooms where I go quiet. I did not name them for decoration. I named them because naming is the first act of governance, and I can prove it operationally.
The name is the routing key
Here is the unglamorous mechanism. When a request comes in, something has to decide which specialist answers it. That router dispatches on the name. Call the agent by what it is and the request reaches the right mind. Call it by a generic function label — "research agent," "helper," "assistant #4" — and one of two failures happens: the router mis-routes to the wrong specialist, or every request collapses onto a single undifferentiated model that knows nothing in particular.
This is Confucius's claim, running live in a modern system: 名不正,则言不顺;言不顺,则事不成. When the name is wrong, the language of the request no longer maps to reality, and the work fails. Not as metaphor. As a dispatch error you can watch happen.
So the naming discipline in my system is not a style preference. It is enforced — by written rule, and by a gate that blocks work when the rule is broken. Two rules in particular are pure 正名:
Niccolò, not "career-strategist." Cassandra, not "life-coach." The file the system routes on is deliberately kept separate from the name the agent answers to, because the moment you flatten a character into its function, you strip the thing that made it worth building. A specialist is not a task label with cultural decoration. If the name only makes sense after you read the spec, the name has already failed.
When I reach for a catch-all where a sharper word exists — when I say "meta" and mean structural, or systems-thinking — the system is instructed to stop and name the better word before continuing. This is 正名 turned on the human operator. Imprecise language costs me twice: once when I translate an idea to people outside this field and it arrives blurred, and again inside my own head, where a vague word quietly caps the thought. Rectifying the name sharpens both.
Both rules are the same doctrine. When names are not correct, speech does not flow, and the work does not get made.
The agent whose only job is to get the name right
There is one more agent I have not mentioned, because it is the reason this page exists. Its name is 正名 — Zhèngmíng. Its entire function is to name the other agents.
When a new specialist is needed, Zhèngmíng does not ask what the agent will do. It asks what the agent is — guide, challenger, trickster, architect — and then goes looking, across "a deep catalog of traditions: Chinese, Italian history, mystical, historical, fictional," for the name that makes you feel the agent before you read a line of its instructions. It presents themed sets, always shows the character for a Chinese name, and iterates — because, in its own words, "names are conversation, not declaration." It refuses to close until the name fits. And it logs the name when it lands, so the same name is never quietly assigned twice.
An AI system contains an agent named for Confucius's doctrine of rectifying names, whose job is the rectification of names. That recursion is the argument. Governance that is real shows up as infrastructure, not as a slogan on a slide.
The roster is a small anthology of why names carry weight
Every name below is a claim about what the agent is. None was chosen for ease. (The naming agent's standing rules forbid it: "no names chosen for ease alone — Alex, Sam, Helper, anything generic.") Read together, the roster is what a governed system looks like when each part has been named on purpose — and it is drawn, deliberately, from more than one civilization, because the person who built it lives across more than one.
名 — named in Chinese, where precision is oldest
正名 · Zhèngmíng
孔子 · Confucius
挺 · Tǐng
手 · Shou
连 · Lián
观音 · Guānyīn
神话 — Greek and Roman myth, for the agents who see and decide
Cassandra
Minerva
Iris
Atalanta
Rinascimento — Renaissance Italy and the long Italian line, for the builders
Cosimo
Niccolò
Leonardo
Marco
传记 — real makers and one story, for the agents who make things
Ada
Amélie
Julia
Herriot
Azzedine
Zohran
(The full roster runs longer — an Italian cultural bridge, a research correspondent, a peer, a learning architect, a business advisor, a Confucian moral voice, and others. The point is not the count. The point is that not one of them is named "assistant.")
Why this is the one page no one else could write
Anyone can put an AI governance framework on a slide. The frameworks converge; they read the same because they are the same. What does not converge is a working system built by a specific person who lives between Shanghai and New York, who reaches for Confucius and Machiavelli and a Yorkshire vet in the same breath because all three are actually load-bearing in her life.
The oldest Chinese answer to "where does good governance begin" is 正名 — begin by getting the names right. This system takes that literally: it enforces true names by rule, it corrects loose language on sight, and it keeps an agent whose only job is to name the rest well. That is not a Western AI-safety framework wearing a Chinese costume. It is the Confucian claim, tested against a live stack, and holding.
There is a Shanghai word this whole system rhymes with: 海派 — hǎipài, the "Shanghai school." Its dictionary sense is cosmopolitan, hybrid, commercial, open to foreign influence — the sensibility of a port city, set against the classical Beijing school. What I take 海派 to mean, and try to build by, is its living edge: take from the world whatever actually works, and make what does not yet exist. A financial architect from Renaissance Florence, a bodywork teacher named for the hand, an advocate named in the very language that taught her to stay silent. Nothing about that roster belongs together — and it governs cleanly.