AI

AI is an umbrella word, not a specific technology. Right now, when someone in a coding context says 'the AI', they almost always mean a large language model wrapped in a harness that lets it read files, run commands and edit code. Ten years ago the same word pointed at recommendation systems and image classifiers; ten years from now it will point at something else. The label follows the frontier.

The vagueness costs you when you're debugging. 'The AI got it wrong' doesn't tell you whether the model produced a bad next-token prediction, the harness fed it a bloated context window, a tool call failed silently, or your prompt was ambiguous. Each of those has a different fix, and the word AI hides which one you're looking at.

The habit worth building is to swap the umbrella word for the specific part. Say model when you mean the trained network, agent when you mean the model plus its loop, harness when you mean the product around it (Claude Code, Codex, Cursor). The rest of this dictionary exists so you have those words ready.