AI Coding Dictionary: A Plain-English Glossary That Names the Parts Others Leave Vague
AI coding jargon, explained in plain English.
At a glance
- What is it?
- Matt Pocock's AI Coding Dictionary translates the jargon of AI-assisted development into plain English, with entries that explain mechanisms, not just definitions. It is a reference for engineers who want to stop guessing at terms like context, harness, and prefix cache.
- Who is it for?
- Adopt this dictionary if you are a developer or team lead who needs a shared vocabulary for AI-assisted coding, especially when onboarding newcomers or writing specs that involve agents and context. Skip it if you want deep technical internals or code-level explanations; the entries are conceptual and deliberately non-academic.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 3 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- mattpocock/dictionary-of-ai-coding5,493 ★ · +1,692
| Repository | 2026-08-18 | 2026-11-15 |
|---|---|---|
| mattpocock/dictionary-of-ai-coding | 3.8k | 5.5k |
1,692 Stars in the last 89 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
What This Dictionary Actually Does
The AI Coding Dictionary is a reference site, not a tool or a library. It defines the vocabulary of AI-assisted coding in plain English, with entries organized into seven sections: The Model, Sessions and Context Windows, Tools and Environment, Failure Modes, Handoffs, Memory and Steering, and Patterns of Work. The README is blunt about the motivation: a lot of the confusion is manufactured, and there is a VC-funded economy that benefits from keeping it hard to understand. The project's bet is that the basic terms are learnable in an afternoon. That bet holds. Each entry reads like a short explainer, not a dictionary stub. For example, the entry for AI itself is a historical table running from 1950s symbolic reasoning to 2020s large language models, ending with the AI effect, where once a technique works reliably it stops being called AI. The dictionary's job is to give you the precise word for the part you mean, so you can say model, harness, or context instead of AI. That is a small but real contribution: it turns vague frustration into testable claims.
The Mechanism: Definitions That Explain Causality, Not Just Meaning
The dictionary does not stop at one-line definitions. Each entry carries a mechanism, a reason why something behaves the way it does. Take the entry on context. The README lists it under Section 2, and the table of contents shows entries like context window, stateful, and agent. The explanations are built around cause and effect: why does context degrade? Why is the bill so high? Why does the same prompt behave differently from one day to the next? The dictionary answers each with a clean answer, once you know the words. This is not a API reference. It is a conceptual map. The value is in the connections: model is stateless, so an agent must supply context via a harness. That causal thread runs through the whole dictionary, and it is what makes the project more than a list of buzzwords.
How It Is Built and How You Run It
The project is a set of Markdown dictionary files plus a small TypeScript script that generates the README. The README states that the content lives in dictionary/*.md, alongside internal/Curriculum.md and internal/README.template.md. The README itself is a generated file: it carries a comment at the top saying GENERATED FILE, DO NOT EDIT, and points to the source files. To regenerate the README, you run npm run generate. That command pulls the Markdown sources and the template, then produces the final README. The README logo links to aihero.dev/ai-coding-dictionary and includes a dark-mode and light-mode logo image. The repository has no GitHub releases. For a developer, the practical takeaway is that the content is fully editable: fork the repo, edit the Markdown files, and run the generator. That is a clean architecture for a dictionary, because it separates content from presentation. You do not need to touch TypeScript code to add a term.
What It Covers: From Prefix Cache to Vibe Coding
The table of contents shows a deliberate scope. Section 1 covers the model itself: parameters, training, inference, tokens, next-token prediction, non-determinism, and the economics of input and output tokens. Section 2 moves to sessions: statelessness, context, context window, agents, system prompts, turns. Section 3 is about the environment: tools, tool calls, MCP, permission requests, sandboxes. Section 4 is failure modes: sycophancy, hallucination, parametric knowledge, attention budget, smart zone. Section 5 covers handoffs between sessions: clearing, compaction, spec, ticket. Section 6 is memory and steering: AGENTS.md, progressive disclosure, context pointers, skills, subagents. Section 7 is work patterns: human-in-the-loop, vibe coding, grilling, DX and AX. That is a broad map, but it is also opinionated. The dictionary picks terms that matter for practical AI-assisted development, not for research. It does not cover model architectures like transformers or attention heads in detail, because the README's audience is developers who use models, not build them. The inclusion of terms like AX (agent experience) suggests a forward-looking stance, but the dictionary stays grounded in what you can act on.
The Limits: A Single Author's View, Not a Standard
The dictionary is authored by Matt Pocock, a well-known TypeScript educator. That is a strength and a weakness. It means the prose is consistent and the explanations are clear, but it also means the terminology is one person's mental model. Terms like smart zone, grilling, and AX are less widely used than terms like token or context window. For an engineer evaluating the dictionary, that matters. If you adopt it as a team reference, you are adopting a specific vocabulary. That vocabulary may not match what your AI vendor calls things. The dictionary is a starting point, not an authoritative spec. You will still need to reconcile its terms with the documentation of whatever tools you actually use.
An Alternative: The Vendor Docs You Already Have
The obvious alternative to this dictionary is the official documentation from the model providers and tool makers. OpenAI, Anthropic, and the MCP spec each publish their own glossaries and conceptual guides. Those docs are authoritative for their own products: they define context window exactly as that vendor implements it, and they give precise pricing for input and output tokens. The difference in approach is fundamental. Vendor docs are product-specific and change with each release; this dictionary is product-agnostic and changes only when Pocock updates it. Vendor docs are written to sell you usage, so they tend to emphasize capabilities and minimize failure modes. This dictionary spends a whole section on failure modes, including hallucination and sycophancy, which are often buried in vendor docs. If you need a neutral overview to understand a concept before reading a vendor's docs, this dictionary works. If you need to know a specific API's limits, go to the vendor. The dictionary does not replace the vendor docs; it prepares you to read them.
Maintenance and License: What You Are Getting Into
The repository shows no license field, which is a red flag for any commercial adoption. The metadata lists it as unknown, and there is no LICENSE file among the top-level entries. Without a license, you have no legal right to copy, modify, or redistribute the content beyond what copyright law grants by default. That is a real constraint if you want to fork the dictionary for internal use. The repository has no GitHub releases, and the last push was on 2026-09-24. The README is generated from source Markdown, so the maintenance model is clear: Pocock updates the dictionary files, and runs npm run generate. There is no CONTRIBUTING file among the top-level entries. The bigger maintenance cost is on your side if you fork it. You will need to track upstream changes to keep your fork current, and you will need to verify that new entries match your team's actual tooling. The project is a static site, so there is no server to run, but the npm generate script implies a Node.js toolchain if you want to build it yourself. For a small team, the cost is low; for a large org, the license uncertainty is disqualifying until clarified.
Editorial conclusion
Adopt this dictionary if you are a developer or team lead who needs a shared vocabulary for AI-assisted coding, especially when onboarding newcomers or writing specs that involve agents and context. Skip it if you want deep technical internals or code-level explanations; the entries are conceptual and deliberately non-academic. Before adopting it as a team standard, verify that the terms match the tools you actually use, since the dictionary is authored by one person and may not cover every vendor's naming. The project is generated from Markdown files, so you can fork it and adjust entries to your stack, but you must accept the maintenance burden of keeping a fork in sync with upstream changes. The dictionary succeeds because it names the parts, so use it to name the parts in your own documentation, then move on.
Frequently asked questions
Dictionary of AI Coding: what are 10 common AI buzzwords that everyone should know?
Ten anchors from the table of contents, each with the bucket it sits in: token and harness (The Model), context window (Sessions, Context Windows and Turns), tool call, MCP, and Sandbox (Tools and Environment), sycophancy, hallucination, and knowledge cutoff (Failure Modes), and compaction (Handoffs).
Dictionary of AI Coding: how is AI actually coded?
As vocabulary rather than as instructions. The entry for AI says the word is a moving label and tells you to swap it for the precise term: the model, the harness, the agent, or the context you gave it. The repository's own code just stitches Markdown together, since `npm run generate` runs `tsx internal/generate-readme.ts` to rebuild README.md from `dictionary/*.md`, `internal/Curriculum.md`, and `internal/README.template.md`.
Dictionary of AI Coding: what language is most AI coded in?
The dictionary's table of contents has no entry on programming languages. What the repository shows is TypeScript as the primary language, a private manifest with type set to module, and tsx ^4.19.2 running the generator alongside typescript ^5.7.2 and @types/node ^22.10.0 in devDependencies.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/mattpocock-dictionary-of-ai-coding)Star history chart for your README
[](https://hysenlabs.com/projects/mattpocock-dictionary-of-ai-coding)Was this useful?