We use cookies to understand how you use this site and improve your experience.

Alexandru Mareș@allemaar
Alexandru Mareș
  1. Home
  2. Writing
  3. The Vocabulary
Email
RSS
YounndAIYou and AI, unifiedBuilt withNollamaNollama

The Vocabulary

Alexandru Mareș

On this page

  • The Vocabulary
  • Named Patterns for AI Reasoning
  • What I Learned from Alexander
  • The Pattern Language → GoF → YON Arc
PreviousThe Pliable Mind
NextBeyond Text: When Agents Speak in Pure Meaning
Related
The Lineage
12/03/2026
The Medium12/03/2026
The Pliable Mind12/03/2026
Published12/03/2026
Read time3 min
Topics
Language & CognitionThinkingPhilosophyYON
Actions
00
Comments

Loading comments…

Leave a comment

0/2000

The Vocabulary

Christopher Alexander published A Pattern Language in 1977, and it changed how architects think.

His argument was deceptively simple: design should use a vocabulary of patterns, reusable solutions to recurring problems at specific scales. A "Light on Two Sides" pattern for rooms. A "Six-Foot Balcony" pattern for outdoor spaces. An "Intimacy Gradient" pattern for building layout.

The patterns weren't just recipes. They were cognitive tools. Each named pattern became a unit of thought that designers could combine, vary, and apply. The vocabulary of patterns determined what the designer could think about. Without a pattern for "Intimacy Gradient," the concept of gradually transitioning from public to private space has to be reinvented every time. With the pattern, it's a building block. Available, named, ready to deploy.

Alexander directly inspired the Gang of Four Design Patterns book (1994), which applied the same idea to software. Singleton, Observer, Factory, Strategy. These patterns became the vocabulary of object-oriented design. Every Java developer thinks in terms of patterns. The vocabulary shapes the design.

What I didn't see until I was deep into YON is that the same mechanism applies to AI cognition.

Named Patterns for AI Reasoning

YON's type system, kind=rule, kind=skill, kind=spec, kind=test, is a pattern language for AI agent behavior.

Each kind is a named pattern. Not just a label, but a cognitive category that shapes how the agent approaches the information:

  • kind=rule → Constraint-oriented reasoning. The agent treats content as enforcement boundaries, things that must or must not happen.
  • kind=skill → Procedural reasoning. The agent treats content as workflows, steps to follow in sequence.
  • kind=spec → Reference reasoning. The agent treats content as ground truth, facts to verify against.
  • kind=test → Verification reasoning. The agent treats content as acceptance criteria, conditions to check.
  • Without these named patterns, the agent processes everything as "text." There's no cognitive distinction between a rule and a description, a skill and a specification. Everything has equal weight and equal cognitive treatment.

    With them, the agent has a vocabulary for types of thought. Just as Alexander's patterns let architects think in terms of spatial concepts, YON's kinds let agents think in terms of behavioral concepts.

    What I Learned from Alexander

    At first I thought YON's type system was a classification tool, a way to organize documents. Then I noticed something in the benchmark data that changed my mind.

    Agents given kind=rule documents enforced the rules more consistently than agents given the same content without the kind marker. The content was identical. The classification created the behavior. The name did the work.

    Alexander would have recognized this instantly. A named pattern doesn't just describe a solution. It makes the solution cognitively available. Before the pattern has a name, you might stumble onto it accidentally. After it has a name, you reach for it deliberately.

    YON's kind system names the patterns of agent behavior. And naming them makes them default. The agent doesn't need to figure out whether something is a rule or a description. The notation tells it. The cognitive work of classification is done by the structure, not by inference.

    The Pattern Language → GoF → YON Arc

    There's an arc here that spans fifty years:

    Alexander (1977): Named spatial patterns create an architectural vocabulary. The vocabulary determines what you can design.

    Gang of Four (1994): Named software patterns create a design vocabulary. The vocabulary determines what you can architect.

    YON (2026): Named cognitive patterns create a behavioral vocabulary for AI agents. The vocabulary determines what the agent can think.

    Each step applies the same insight to a more abstract domain. Architecture to software to cognition. Spaces to code to minds.

    I'm not sure where the next step goes. But if the pattern holds, that named vocabularies create cognitive capabilities in whatever domain they're applied to, then the design of YON's type system is one of the most consequential decisions in the project. Not because of what it classifies, but because of what it enables the agent to think.


    Previous: The Lineage Next in the series: The Warning, on what happens when constructed languages create realities.