Glossary

LLM — Large Language Model

While there are other approaches to creating artificial intelligence (A.I.), the technology sweeping the world as of 2026 heavily relies on Large Language Models, or LLMs. In its simplest form, an LLM is a computer program that converts vast amounts of written content into numbers called "tokens." The program then uses massive amounts of computer processing power, which allows it to use these tokens to make predictions about which words should come next in a sentence. When combined with other technologies, these predictions can seem almost human. This core technology is extended to create images and videos and, increasingly, to make complex, real-world decisions.

ChatGPT, Claude, and Gemini are all built with an LLM at their core.

Agent

A chatbot answers one question at a time. An agent is given a goal and works toward it on its own: it decides on a step, carries it out, looks at the result, and decides what to do next until the job is done. That's what lets it book a table, compare prices across websites, or fix a piece of software while you do something else. It's also where mistakes cost the most — an agent acting on a wrong guess can do real damage before anyone notices.

Claude Code and OpenAI's Operator are both agents.

Tool

On its own, an LLM can only read and write text. A tool is a specific action it has been given permission to take: search the web, open a file, run a calculation, send an email. The more tools a chatbot has, the more it can actually do for you — and the more it's being trusted with.

When ChatGPT searches the web or Claude reads a spreadsheet you attached, a tool is doing the work.

Skill

A skill is a set of instructions for one kind of task, written down once so a chatbot like ChatGPT or Claude can follow it every time — how to research a company, how to write a cold email, how to format the monthly report. Without one, you'd explain your preferences from scratch in every conversation. A growing library of skills is how a chatbot gets better at the work you actually do.

MCP — Model Context Protocol

Chatbots like ChatGPT, Claude, Gemini, and others have the ability to communicate with outside websites and other applications. Model Context Protocol is a fancy-sounding term that is basically a standardized language which facilitates this communication. ChatGPT, for example, would use MCP to reliably get flight information from Google, or transactions from your bank… if you've allowed it.

RAG — Retrieval-Augmented Generation

An LLM only knows what it read during training, and that reading stopped at some point in the past. RAG gives it fresh, specific material at the moment you ask: the program first searches a set of documents — a company's files, medical guidelines, last week's meeting notes — then hands the most relevant passages to the LLM to answer from. The answer comes from your material rather than the LLM's memory, so it's more current and far less likely to make things up.

When a company's help chatbot answers from its own manuals, RAG is usually what's happening.