The VC-funded fog around AI coding, decoded

What it does This repository is a curated dictionary of roughly 50 AI coding terms—from “next-token prediction” to “vibe coding”—each defined in plain English and paired with realistic dialogue showing how to use (and avoid misusing) the word. It is generated from Markdown sources and published as a readable reference. Think of it as a phrasebook for the conversation happening in your terminal.

The interesting bit The project argues that much AI coding confusion is manufactured by a “VC-funded economy” that benefits from opacity. The antidote is not another framework, but vocabulary: once you know the difference between a model, a harness, and a provider request, the “mysterious failures” and inexplicable bills become legible.

Key highlights

• Covers seven domains: model internals, context windows, tools, failure modes, handoffs, memory, and workflow patterns.
• Each entry includes a concise definition and a usage snippet in dialogue format, which makes the terms stick better than a wall of text.
• Explicitly distinguishes concepts often conflated, like training vs. inference or input tokens vs. output tokens, with billing implications noted.
• Defines operational terms like prefix cache and attention degradation that directly affect cost and code quality.
• Generated from source Markdown files via a build script, so it is maintainable as a structured reference.

Caveats

• It is a glossary, not a tool: the TypeScript scaffolding is just build glue for a documentation site.
• The README is the entire product; there is no interactive search or API.
• Some entries are truncated in the provided source (e.g., Prefix cache), though the full content likely exists in the source Markdown.

Verdict Read this if you are using Claude Code, Cursor, or similar tools and feel like you are “vibe coding” through a fog of unexplained bills and erratic behavior. Skip it if you are already fluent in tokens, harnesses, and context-window management.