AI Coding Dictionary: Plain-English AI Coding Jargon
/dictionary-of-ai-coding
AI Coding Dictionary is a plain-English glossary that translates the jargon of AI-assisted development into concepts any developer can reason about. It covers how LLMs actually work — tokens, inference, context windows — without hype or marketing language. Reach for it when a tool's behaviour seems arbitrary or a bill doesn't match the work; skip it if you want hands-on tutorials or framework-specific recipes.
Why AI Coding Feels Confusing
AI coding vocabulary is opaque by design. The README explicitly names 'an economy, VC-funded, that gains from making understanding difficult.' Terms like 'context window', 'inference', and 'hallucination' get used without explanation, and the same words mean different things depending on who says them — the README traces how 'AI' alone has meant expert systems, deep learning, and large language models at different times. Without the vocabulary, debugging a puzzling model behaviour or reading a token bill becomes guesswork.
What the Dictionary Covers
AI Coding Dictionary is a curated glossary aimed at developers who find AI-assisted coding frustrating or opaque. It translates terms like hallucination, context window, and inference into plain-English explanations grounded in how the underlying mechanisms actually work. Each entry explains the mechanism behind the behaviour, with concrete usage examples showing when to care about each concept. It assumes no machine learning background.
Key Concepts Explained
- ✓Token and context window mechanics — why code with unusual identifiers consumes more context than it looks like it should
- ✓Model vs. harness distinction — a model is stateless next-token prediction; everything agentic lives in the harness around it
- ✓Training vs. inference lifecycle — parameters are frozen after training, which explains why corrections don't stick and knowledge has a cutoff date
- ✓Non-determinism and its consequences — same prompt, different output; retry as a legitimate strategy; why automated checks matter more than with deterministic tools
- ✓Effort levels and when to use them — low for mechanical edits, high for concurrency bugs and design decisions, with guidance on matching effort to task
- ✓Failure modes — hallucination, context degradation, and why they happen mechanically rather than as mysterious malfunctions
- ✓Handoffs and session management — how context carries across turns and why it degrades over time
Strengths
- ✓Plain-English explanations that demystify jargon without dumbing it down — the 'AI' entry alone is worth reading to understand why AI conversations talk past each other
- ✓Mechanistic explanations rather than just definitions — you understand why a behaviour happens, not just what it's called
- ✓Strong on foundational concepts — tokens, inference, context windows, and parameters are covered with enough depth to actually reason with them, not just nod at
- ✓Usage examples in every entry showing how to apply the concept in conversation — translates vague complaints into precise specs
Common Misconceptions and Nuances
- △Static reference — no interactive examples, quizzes, or hands-on exercises; it's a glossary you read, not a course you complete
- △Doesn't cover specific framework APIs or tool integrations — won't tell you how Cursor, Copilot, or any specific harness actually works in practice
- △No version history or update timestamps visible — when a concept's meaning shifts (as 'AI' has repeatedly), there's no signal for how current each entry is
- △Covers concepts but not workflows — you'll understand the vocabulary of AI coding after reading it, but not necessarily how to build an agentic pipeline with that vocabulary
Frequently Asked Questions
AI Coding Dictionary translates AI coding jargon into plain English for developers who find the vocabulary opaque. Each entry explains not just what a term means but why the underlying mechanism produces the behaviour it describes.
Seven sections cover models, context windows, tools and environment, failure modes, handoffs, memory and steering, and patterns of work. Core concepts include tokens, parameters, inference, effort levels, and non-determinism.
The dictionary calls 'AI' a moving label that points at whatever computers can newly impressively do. It traces what the term has meant across decades — expert systems, deep learning, large language models — to show why technical claims using the word carry a hidden timestamp.
Same prompt, different output. Because models sample from a probability distribution, retrying is a legitimate strategy when a response is poor. This also means you cannot test an agent once and rely on repeatability — automated checks catch bad draws that manual testing misses.
The AI Coding Dictionary links to the aihero.dev newsletter, which has over 62,000 subscribers. The README directs readers there for ongoing thinking on AI engineering beyond the glossary vocabulary.
The repository metadata does not clearly state a license. Before reusing or redistributing the content, verify the license status on the GitHub repository page, as unspecified licensing may restrict commercial use.
Alternatives
Best use cases
- •Debugging a model that confidently gives wrong answers — the dictionary explains hallucination mechanistically so you stop treating it as a personality flaw
- •Reading a token bill that doesn't match your expectations — understanding that cost scales with context size, not file count, changes how you structure prompts
- •Onboarding a junior developer onto an AI-assisted team — send them here before they ask 'why did it ignore my correction?'
- •Choosing between model tiers mid-session — knowing that the harness does most agentic work helps you decide when a smarter model actually matters
- •Understanding why the same prompt behaves differently on different days — non-determinism and provider-side batching are real, not imaginary
Who should try it — and who should skip
Reach for AI Coding Dictionary when you are a developer working with LLM-powered tools but don't have a machine learning background — you can learn the basic terms in an afternoon and stop feeling like you are guessing. It is also useful for technical leads who need to communicate precisely about AI costs, failure modes, or architecture decisions without vendor marketing clouding the conversation. Skip it if you already understand training versus inference, what a token is, and why the same prompt can produce different output — you have moved past vocabulary and into workflow. Also skip if you want framework-specific integration guides, because this covers concepts, not code.
Still deciding about dictionary-of-ai-coding?
One click hands the question to an AI along with this page — see what it says about dictionary-of-ai-coding.