We use cookies to understand how you use this site and improve your experience.
I've spent six essays exploring other people's ideas. Whorf's grooves, McLuhan's media, Ong's history, Alexander's patterns, Borges's warning, and the pliability thesis that ties them together.
This essay is different. This one is mine.
YON started as a practical problem. I needed AI agents to follow behavioral contracts, sets of rules about how to interact with specific parts of a codebase. Markdown worked sometimes. JSON worked sometimes. Nothing worked reliably.
So I built a notation. I gave it typed tags, scoped sections, enforcement levels, density profiles. I colocated it with source code in sidecar manifest files. I designed it to be line-oriented and streamable, because agents process text in streams, and a format that requires full parsing before comprehension is a format that fights the medium.
I didn't start with philosophy. I started with compliance rates. And the compliance rates improved. Not marginally. From 57% to 85% on first-attempt adherence. The same instructions, restructured as notation, produced fundamentally different agent behavior.
That's when I started reading Whorf.
Over time, the practical decisions crystallized into four principles. I wrote them into The Guide, a document that comes before the specification, because philosophy must precede architecture.
Discipline with Flow. Every act of creation balances structure and intuition. YON's grammar is strict: typed tags, explicit keys, enforcement levels. But the values are free-form. Write what you mean. The structure holds the intent; the flow carries the humanity.
Human before Machine. AI extends human potential. It doesn't replace it. YON encodes human intent as human intent. It doesn't translate or rephrase or optimize away the original meaning. The format preserves the author's voice.
Structure before Scale. Complexity is earned through clarity. YON starts with single-line declarations before it allows nesting. It starts with three profiles before it allows extension. Simplicity first, complexity when justified.
Continuity before Chaos. Memory sustains meaning. Context sustains truth. YON is streaming-first and fault-isolated. If the parser hits a malformed line, it recovers on the next line, not the next document. Partial is valid. The stream doesn't break.
These aren't abstract values. They're engineering constraints that shaped every syntax decision in the format. The philosophy is the architecture.
Looking back through the lens of this series:
Whorf gave me the grooves, the understanding that notation creates default cognitive pathways. I stopped thinking of YON as a container and started thinking of it as a landscape.
McLuhan gave me the distinction between storage, display, and cognitive media. I stopped optimizing YON for readability and started optimizing it for reasoning quality.
Ong and Goody gave me the lineage, the realization that notation systems have always created cognitive capabilities. Writing created logic. Mathematics created proof. YON creates something I'm still learning to name.
Alexander gave me the pattern language, the understanding that named patterns become cognitive primitives. kind=rule isn't a label. It's a building block of agent thought.
Borges gave me the warning, the knowledge that constructed languages don't just describe worlds, they create them. YON doesn't just document systems. It constructs the cognitive reality agents inhabit.
The pliability thesis gave me the urgency. LLMs are pliable in a way humans aren't. The window for deliberate cognitive engineering through notation design is open now. I don't know how long it stays open.
I believe the weak Sapir-Whorf form applies to YON and LLMs. I think the strong form might apply. I can't prove either definitively.
I believe notation design is cognitive architecture. I think it's underappreciated. I can't prove it matters at scale. Not yet. The benchmarks are suggestive, not conclusive.
I believe the YON type system creates cognitive primitives for AI agents. I think kind=rule does something fundamentally different from a JSON key called "type." I can see it in the compliance data. I can't fully explain the mechanism.
There was a time when I mistook uncertainty for weakness. I don't anymore. It might be the most honest thing I have to offer. The questions are better than the answers. And the questions are what keep me building.
There's a name for what I built. I call it The Quiet Law.
It's quiet because the best notation disappears. You don't notice the grammar, you notice the clarity. You don't see the structure, you see the intent. The format steps back so the meaning can step forward.
It's law because the structure isn't optional. Not "best practice." Not "when possible." The typed tags, the enforcement levels, the scoped sections: they're the grammar of a cognitive medium. You speak it or you don't. There's no in between.
The format is the format because the philosophy required it.
The Guide declares. The Codex translates. The Words speak. The Sight shows. The Mind explains why.
And this series, these seven essays, is my attempt to explain the "why" not as a technologist, but as a person who started thinking about notation and ended up thinking about minds.
Previous: The Pliable Mind This is the final essay in the "Designing Languages for Artificial Minds" series. Start from the beginning: The Grooves.
Loading comments…