The Power of Naming: Key Insights in AI Prompt Engineering

In Patrick Rothfuss's The Kingkiller Chronicle, the legendary wizard Taborlin the Great draws his power from something remarkably simple—he knows the true names of all things, and thereby commands them. He calls the name of stone, and stone breaks apart. He leaps from a high tower, calls the name of wind, and the wind bears him safely to the ground.

This literary metaphor precisely captures the core dilemma of today's AI prompt engineering: the vaguer you are, the more easily the AI drifts from your true intent.

Why AI Always Seems to "Misunderstand" You

The image in your mind is crystal clear, yet the generated result looks like an artistic disaster. The feature requirements seem simple and straightforward, but the implementation veers completely off course. More frustrating still, you can't even articulate what exactly is wrong with the first version.

The author compares this experience to wine tasting: a connoisseur can describe in detail the balanced, multi-layered complexity of a wine, while an ordinary person can only say "I like this one" without being able to explain why.

This gap in expressive ability is precisely what causes the enormous divergence in AI experience among different users.

"Vibe Coding": An Overloaded Term

The debate about "vibe coding" is essentially a confusion between two fundamentally different practices.

One interpretation: prompting AI entirely by feel, with no engineering rigor whatsoever—this is the pejorative reading of vibe coding. The other interpretation: using natural language to direct AI work rather than hand-writing syntax line by line—this is a neutral, even positive description of vibe coding.

In engineering practice, you often hear complaints like: "The code architecture AI generates is just garbage." Such judgments stem precisely from conflating these two interpretations—the problem isn't using natural language to guide AI, but failing to anchor architectural constraints with precise terminology.

The Qualitative Shift from Precise Naming

Let's look at a concrete engineering example.

If you tell AI: "Implement code that fetches data from service A, transforms it, and loads it into the data warehouse", you'll likely get a 500-line monster method with all logic piled together, direct inline database calls, no abstraction layers, and no tracking of historical changes.

But if you rephrase it—telling it to isolate side effects behind DAO and service client classes, use IoC and DI to ensure testability, organize the repository structure around a medallion data model, and employ Type 2 SCDs to track historical changes—you'll get a dramatically different quality of result, with virtually negligible extra effort.

The difference is simply this: whether you know the names of these fundamental patterns, and when to use them.

This naming ability is not limited to the coding domain.

DAO (Data Access Object) is a design pattern whose core idea is to encapsulate all database access operations in dedicated classes, so that the business logic layer need not concern itself with underlying storage details. The value: when you need to switch databases, add caching, or modify query logic, you only change the DAO layer without affecting upper-layer business code.

IoC (Inversion of Control) and DI (Dependency Injection) are a tightly related pair of concepts. IoC refers to transferring the creation and lifecycle management of objects from within the code to an external container; DI is the concrete means of achieving IoC—"injecting" dependency objects into consumers via constructor parameters or property assignment. Together, the effect is that classes no longer hard-code their dependencies, making it easy to swap in mock implementations during testing.

Medallion data model is a layered data architecture, typically divided into bronze (raw data), silver (cleaned and transformed data), and gold (business-oriented aggregated data) layers. It organizes the data warehouse through progressive refinement, making data lineage clear and quality controllable.

Type 2 SCDs (Slowly Changing Dimensions Type 2) is a classic technique in data warehousing for tracking historical changes: when a dimension record changes, rather than overwriting the old value, a new row is added with a marked effective time interval, preserving the complete historical version.

The Cross-Domain Transfer Value of Naming Ability

Image generation follows the same principle. If you want to describe an artistic style but don't know its name, you might only say "old-fashioned storybook style" and then pray the model shares your understanding. But when you learn that the effect you want is called Art Nouveau, you gain a far more precise starting point.

The same applies to technical documentation writing. Rather than vaguely asking to "keep it concise and clear," directly specify writing in Simplified Technical English—the latter's constraints and operability far exceed the former.

The author admits that they now spend more time learning the fundamentals of unfamiliar domains: studying art styles to better generate images, studying writing styles to generate text suited for specific purposes. And AI itself has become the best learning tool—you can simply throw an image in and ask about its artistic style, something previously unimaginable.

Art Nouveau is a decorative art style that flourished in Europe from the late 19th to early 20th century. Its typical features include: extensive use of organic curves from plant vines and flowers, emphasis on flowing movement and asymmetrical composition, commonly seen in architecture (such as Paris Metro entrances), posters (such as Mucha's works), and jewelry design. Using this term directly in image generation prompts allows the model to accurately target specific color tendencies, line styles, and compositional logic—far superior to vague descriptions like "old-fashioned storybook style."

Simplified Technical English (STE) is a controlled natural language standard, originally developed by the aviation industry for writing maintenance manuals and other technical documents. It strictly limits the permitted vocabulary (approximately 900 core words), sentence structures, and sentence length (typically no more than 20 words), and requires each word to express only one meaning. When you specify "write in STE style" in a prompt, the model will automatically follow these constraints, producing highly consistent, unambiguous text.

Semantic Anchors: A Practical Starting Point for Prompt Engineering

If you're looking for an entry point to systematically learn these "true names," the Semantic Anchors project is worth attention. This open-source project attempts to collect and organize a large number of pattern, style, and concept names that are extremely useful in prompts.

Taborlin the Great gained power by knowing the true names of all things; AI grants us the same superpower at a much lower cost. Rather than complaining that AI-generated code architecture is terrible or that image styles are wrong, spend time expanding your "naming vocabulary"—this is precisely the most underrated core skill of the AI era.

One-line Summary

Knowing the precise "true names" of things is the core skill of AI prompt engineering; naming ability transfers across domains and can significantly improve output quality.

Paragraph Summary

Opening with the literary metaphor of Taborlin the Great from The Kingkiller Chronicle who gains power by knowing the true names of all things, the article points out that the core dilemma of AI prompt engineering lies in expressive precision: the vaguer you are, the more easily AI drifts from your true intent. The term "vibe coding" is overloaded, conflating "casual prompting" with "effectively guiding AI using natural language." Through coding examples, the author shows that specifying precise terminology such as DAO, IoC, DI, medallion data model, and Type 2 SCDs enables AI to generate dramatically different quality code at virtually negligible extra cost. This principle applies equally to image generation (Art Nouveau) and document writing (Simplified Technical English). The author recommends proactively learning fundamental terminology across domains and suggests the Semantic Anchors open-source project as a systematic starting point, arguing that expanding your "naming vocabulary" is the most underrated core skill of the AI era.