The Vibe Coding Glossary

Building software by describing what you want in natural language and letting AI write the code. The vibe is the intuition; the code is the byproduct.

The instruction you give the AI. A good prompt is clear, specific, and gives context. See Prompting for Vibe Coders.

Large Language Model — the AI engine that predicts text. It's the brain behind the code generation, but it has no memory of your project unless you feed it back in.

A chunk of text the model processes — roughly a word or part of one. Models are billed and limited by token count, so bigger isn't always better.

How much text the model can "see" at once. Once you exceed it, older context falls off — which is why the AI sometimes forgets earlier instructions.

An AI that can take actions — not just answer, but run commands, edit files, and loop until a task is done. More powerful, and more able to make a mess.

The background info you give the AI so it understands your project: existing code, goals, constraints. Good context is the single biggest lever on output quality.

Model Context Protocol — a standard way to connect an AI to external tools and data sources so it can act on the real world, not just talk about it.

Where you write code. Modern AI editors (like Cursor) bake an assistant directly into the editor. See The Vibe Coder's Toolkit.

A view of what changed — line by line — between two versions of code. Always read the diff before accepting an AI's edits.

Saving a snapshot of your code with version control (Git). Every commit is a save point you can return to when a prompt goes sideways.

When the AI states something false with total confidence — a fake library, a made-up API. Always verify before trusting. See Mistakes & How to Survive Them.

When a small project balloons into a sprawling mess because the AI happily keeps adding features. The model never says "that's enough" — you have to.

A password-like credential (database URL, API token). Never hardcode these in source files — keep them in environment variables so they don't leak.

Asking the AI to "clean everything up" and getting back a broken codebase. Prefer targeted fixes over wholesale rewrites.

Code that runs locally but breaks when deployed. The gap between "it works here" and "it works everywhere" is the whole job of shipping. See the shipping guide.