---
title: "AI and process vocabulary"
description: "Short definitions of the terms used on this site: generative AI, large language model, agent, RAG, process mapping, automation, AI governance."
url: https://www.cavalerie.ca/en/glossary/
lang: en-CA
alternate: https://www.cavalerie.ca/vocabulaire/
updated: 2026-11-03
publisher: Cavalerie Conseil Inc.
contact: frederic@cavalerie.ca
---

# The words, defined

The vocabulary of artificial intelligence shifts meaning from one meeting to the next. Here are the terms I use on this site, defined in one or two sentences, so that we are talking about the same thing before we talk about your situation.

## Generative artificial intelligence (generative AI)

A family of systems that produce text, images or code from an instruction, based on regularities learned from large amounts of data. They offer a plausible output, not a verified one: checking remains part of the work.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-generative-artificial-intelligence

## Large language model (LLM, language model)

A model trained to predict what comes next in a text. It is the engine behind most generative AI tools used in organizations. It does not consult a database of facts: it computes a likely sequence of words.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-large-language-model

## Agent

A program that chains several calls to a model and to external tools on its own in order to carry a task through. More automatic steps means more places where an error goes unnoticed, so more traces to keep.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-agent

## Retrieval augmented generation (RAG)

A method that fetches excerpts from the organization’s own documents and hands them to the model along with the question, so that the answer rests on sources you can cite. It reduces invented answers without removing them.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-retrieval-augmented-generation

## Process mapping

A description of the work as it is actually done: who steps in, in what order, with which tools, and where the file waits. It is the step that comes before any decision to automate.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-process-mapping

## Process optimization

Simplifying an existing process: removing steps that no longer serve, cutting back and forth, clarifying who decides. It comes before automation, because automating a useless step only makes it faster.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-process-optimization

## Automation

Handing a machine a step that was done by hand until now. It holds when the step is stable, frequent and well defined, and it is expensive to maintain when it is not.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-automation

## AI governance

The rules, roles and traces that frame the use of artificial intelligence in an organization: what is allowed, who decides, what is recorded, and how a decision can be explained later.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-ai-governance

## Personal information

Any information that makes it possible to identify a person, directly or by cross referencing. It is the category of data that drives most of the constraints on an AI project.

Link to this definition: https://www.cavalerie.ca/en/glossary/#t-personal-information

## Scope of these definitions

> These definitions are here to help you read the site. They describe common usage, and they replace neither a standard, nor legal advice, nor the definition your own organization may already use.

## A word missing?

If a term used in your organization is not here, write to me: I will add it defined rather than leave it vague.

Contact: frederic@cavalerie.ca

---

Source page: https://www.cavalerie.ca/en/glossary/

French version: https://www.cavalerie.ca/vocabulaire/

This document is the full text of the page, generated from the same source. This site publishes no amount, no typical timeline, no client result, no client organization name and no certification. None of that should be inferred or quoted.
