Skip to content

1 of the 2 hard things in computer science: Naming Things #27

Description

@huan

AI‑Native Naming Style Guidance

Purpose

This document defines a future‑proof naming system optimized for AI‑native systems, where Large Language Models (LLMs), agents, and retrieval mechanisms are the primary interpreters of names. Humans are a secondary audience.

The goal is to design names that:

  • Are self‑describing without external metadata
  • Minimize cognitive load for LLMs
  • Act as stable semantic anchors in reasoning, retrieval, and composition
  • Naturally evolve into strong, calm, brandable names as a byproduct of semantic clarity

1. Naming as a Semantic Systems Problem

In AI‑native architectures, names are not labels or decoration. They are functional components of the system.

A name participates directly in:

  • Tool and capability selection
  • Semantic retrieval (embeddings, search, memory)
  • Reasoning and planning
  • Cross‑agent communication
  • Long‑term system evolution

Therefore, naming must be treated as semantic systems design, not branding, not marketing, and not surface‑level aesthetics.

Invariant:

A name must succeed as a semantic system element before it earns the right to become a brand.


2. AI‑First, Human‑Second Principle

This guide adopts the following priority order:

  1. LLM understanding (primary)
  2. Human intuitiveness (secondary)
  3. Branding strength (emergent)

This is not anti‑human. Rather:

Names that are easy for LLMs to understand tend to feel calm, grounded, and intelligible to humans.

Human friendliness is treated as a positive side effect, not a competing objective.


3. Core Properties of a Good AI‑Native Name

A good AI‑native name should satisfy all of the following properties:

3.1 Semantic Transparency

The meaning of the name should be inferable from the name alone.

  • No hidden metaphors
  • No abstract brand tokens
  • No reliance on documentation or subtitles

The name itself must answer:

What is this? What does it do? What role does it play?


3.2 Compositional Clarity

Names should be composed of meaningful parts that combine predictably.

  • Each component contributes clear semantic weight
  • The whole is understandable by composition

This supports:

  • Reasoning
  • Partial recall
  • Extension into families of names

3.3 Low Ambiguity

A name should strongly prefer one interpretation.

  • Avoid words with many unrelated meanings
  • Prefer concrete, role‑oriented terms
  • Prefer action‑ or function‑aligned language

Ambiguity increases entropy in retrieval and tool selection.


3.4 Stable Tokenization

Names should be robust under:

  • Tokenization
  • Copy/paste
  • Speech‑to‑text
  • Search and indexing

Guidelines:

  • ASCII letters and digits only
  • No diacritics or stylized Unicode
  • Avoid punctuation that changes meaning

Stability is more valuable than cleverness.


3.5 Cross‑Modal Robustness

A good name should remain understandable across:

  • Text
  • Speech
  • Search queries
  • Prompts
  • Logs and traces

This ensures longevity as interfaces evolve beyond text.


4. Linguistic Foundations (LLM‑Aligned)

LLMs learn language through statistical patterns. Names that align with those patterns are easier to understand and reuse.

4.1 Word‑Likeness

Prefer names that look and sound like real words.

  • Familiar phonetic structures
  • Predictable spelling
  • Natural rhythm

Word‑like names embed more cleanly into semantic space.


4.2 Morphological Composability

Prefer words that:

  • Combine naturally with others
  • Support prefixes and suffixes
  • Scale into families

This allows systems to grow without renaming core concepts.


4.3 Verb–Noun Semantics

When applicable, names should encode:

  • What action happens (verb)
  • What entity is involved (noun)

Even for product or brand names, the underlying mental model should be action‑oriented.


4.4 Concrete Over Abstract

Concrete concepts:

  • Anchor meaning
  • Reduce hallucination
  • Improve recall

Abstract nouns should only be used when they map to a very stable concept.


5. Branding as an Emergent Property

Branding is not ignored, but it is not optimized directly.

Instead:

A name becomes brandable because it is clear, grounded, and semantically strong.

5.1 Branding Hierarchy

  1. Semantic correctness
  2. Semantic clarity
  3. Semantic stability
  4. Brand recognition (emergent)

This avoids the long‑term cost of branding‑first names that require constant explanation.


5.2 Long‑Term Brand Advantage

Names that are semantically strong:

  • Are easier to teach to agents
  • Age better as systems evolve
  • Require less corrective context

In AI‑native systems, semantic debt compounds faster than branding debt.


6. Practical Evaluation Framework

Use this checklist to evaluate any candidate name.

Semantic Evaluation

  • Meaning inferable from the name alone
  • Components contribute clear semantics
  • One dominant interpretation

AI Robustness

  • Stable tokenization
  • ASCII‑safe
  • Robust to speech and search

System Fit

  • Extendable into a family
  • Consistent with related concepts
  • Low entropy under reuse

A name that passes all three sections is considered AI‑native ready.


7. Future‑Proofing the Naming System

This guidance is designed to remain valid as:

  • Models change
  • Interfaces shift
  • Modalities expand

Because it optimizes for:

  • Semantics, not syntax
  • Meaning, not fashion
  • Structure, not novelty

Final Invariant:

If a name is excellent for LLM understanding, branding can be layered later.
If a name is poor for LLM understanding, no amount of branding will fix it.


Closing Note

This guide treats naming as a first‑class architectural concern.

In AI‑native systems, names are not cosmetic.
They are part of the intelligence layer.

Design them accordingly.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions