// the find
mattpocock/dictionary-of-ai-coding
AI coding jargon, explained in plain English.
A glossary of AI coding terminology written by Matt Pocock (Total TypeScript), aimed at developers who use LLM-based tools daily but trip over terms like 'context window', 'inference', or 'prefix cache'. Each entry goes deeper than a one-liner — 'Token' actually explains tokenization mechanics; 'Non-determinism' explains why you shouldn't over-interpret a string of bad days. This is documentation, not a library.
The depth-to-length ratio is good: entries are long enough to explain the mechanism, not just the definition — 'Cache tokens' includes a worked request/prefix table that most people who hit billing surprises would benefit from reading. The internal cross-linking between terms is thorough and navigable; following [harness] → [tool] → [tool call] reads like a coherent mental model, not disconnected entries. The README is generated from individual Markdown files per term, which is the right call — it means contributors can add/edit one entry without touching a 3000-line file and git blame stays useful. The framing ('a whole VC-funded economy that benefits from keeping it hard to understand') sets the right adversarial tone and earned the stars.
The TypeScript label is slightly misleading — this is almost entirely a docs/Markdown repo with a small generation script; there is no library to install or API to call, which some star-gazers will find out the hard way. The glossary stops at agent-loop concepts and has no entries for anything below that layer (embeddings, RAG, fine-tuning, quantization), so it covers 'how to use Claude Code' well but leaves out half of what a developer building AI features actually needs to know. No versioning strategy for terms — 'MCP' and 'AGENTS.md' are fast-moving targets and there's no indication of when an entry was last reviewed or against which harness version it was written.