Most people who use AI focus on the wrong thing. They chase prompt templates, copy-paste frameworks, and hunt for the perfect instruction set that will unlock some hidden capability. They are optimising the input when they should be upgrading the operator.

The prompt trap

Prompts are important. But treating prompt engineering as the core skill of AI fluency is like treating gear selection as the core skill of driving. It is necessary, it matters at the margins, and it is not what separates good drivers from bad ones.

What separates AI practitioners from AI passengers is not their prompts. It is their mental models: the internal representations they hold of what the technology is, how it works, and what it can do.

A mental model is not a technical specification. It is a working theory that lets you make good decisions quickly, even when the situation is novel.

What a mental model does

A good mental model of AI lets you:

  • Predict where it will succeed and where it will fail
  • Decompose complex tasks into components the model can handle
  • Recognise when output quality is degrading and diagnose why
  • Choose the right model, temperature, and approach for a given task
  • Recover from failures without starting over

None of these come from memorising prompt syntax. They come from understanding the technology at a conceptual level.

12
Mental models covered in Artificial Leverage
Source: Artificial Leverage, Chapter 1 →

The practitioner advantage

Practitioners iterate. They treat every interaction as an experiment, adjust their approach based on results, and build intuition over hundreds of hours of deliberate use. They do not need a template because they understand the principles behind the template.

This is why the book focuses on mental models rather than prompt libraries. A prompt library gives you fish. Mental models teach you to fish, build a boat, and map the ocean.

The gap between knowing AI exists and knowing how to think with it is the most consequential skills gap of this decade.

Artificial Leverage, Introduction

Where to start

If you are new to AI, start with three foundational models:

  1. The probabilistic parrot model. Understand that language models predict tokens, not truth. This single insight prevents most beginner mistakes.
  2. The context window model. Think of the model's working memory as a fixed-size desk. Everything on the desk influences the output. Managing what goes on that desk is half the skill.
  3. The delegation model. Treat the AI as a capable but literal junior employee. Be specific about what you want, provide context, and review the work.

These three models alone will put you ahead of 90% of users who are still copying prompts from Twitter threads.

Mental models compound. Each one you internalise makes the next one easier to learn and apply.