mattpocock/dictionary-of-ai-coding

What it solves

AI coding tools often feel mysterious, expensive, and prone to unpredictable failures. This project demystifies the field by translating the jargon used by AI companies into plain, actionable English, helping developers understand why models behave the way they do and how to use them effectively.

How it works

The dictionary breaks down the vocabulary of AI engineering into seven logical sections: The Model, Sessions, Tools, Failure Modes, Handoffs, Memory, and Patterns of Work. It explains core concepts—like how models are stateless, why they hallucinate, and how they are billed—by focusing on the underlying mechanics of next-token prediction and context management rather than marketing hype.

Who it’s for

This is for developers who want to move past "vibe coding" and guesswork. It is designed for anyone who uses AI coding tools and wants to understand the technical "why" behind their performance, costs, and limitations.

Highlights

• Plain-English translations: Replaces confusing industry jargon with clear, technical explanations.
• Mechanistic focus: Explains how models actually work (e.g., parameters, inference, and tokens) so you can diagnose issues yourself.
• Practical guidance: Provides "Usage" examples for each term to show how to apply the knowledge in real-world development.
• Structured learning: Organizes terms by category, from the basics of the model to advanced patterns like agentic workflows and human-in-the-loop systems.