Here are thirteen words you will hear about AI at work. Each one gets a plain meaning and a place you will meet it. Your task is to write your own version and get a friend to find the fuzzy one.

Learning these is like learning the tool names in a kitchen before you try a recipe. Nobody explains what a whisk is halfway through beating eggs. It is also like a new coworker's first week, spent quietly learning what the office means by "the deck" and "the client." And it is like a phrasebook before a trip abroad. You will not need every phrase on day one, but the few you do need will save you.

Sam ran into this in his first economics lecture, when the professor said "context window" and "hallucinated citation" in the same sentence, and half the room nodded like they understood. They did not. Maria's cousin, who works in software, talks about "agents" and "RAG" over dinner, and Maria used to just smile and change the subject. A sheet like this one turns that smile into an actual question.

Word	What it means	Where you meet it
Model	The trained thing inside the app that turns your words into an answer. Claude, GPT, and Gemini are families of models.	The picker in any AI app where you choose one.
Token	A small chunk of text, about one word or part of a word. The AI reads and writes in tokens.	Price pages and limits, which count tokens.
Context window	How much the AI can look at at once. Like a table that fills up. When it is full, the oldest things fall off.	Long chats that start to forget.
Prompt	Whatever you give the AI to work from. Your question, your notes, your example. Your ask.	Every message you send.
System prompt	Standing instructions the app or you put in front of every chat. It tells the AI who it is and how to act.	Custom instructions. Project rules. The rules inside a company chatbot.
Hallucination	A sure-sounding answer that is not true. It happened because it was the most likely next words.	Made-up book titles. Wrong numbers. Fake features.
Fine-tuning	Extra training with chosen examples to change how a model acts.	Company news. A business training a model on its own files.
Retrieval (RAG)	Finding the right bits of your own files and handing them to the AI before it answers. RAG is the short name.	Company chatbots that answer from files. Job ads.
Agent	An AI set up to do a job in steps, using tools, not just answer once. "Agentic" means built that way.	Coding agents. Browser agents. Job ads. Modules 3 and 4.
Tool	Something the AI can use to do a thing. Search. A calculator. Read a file. Send an email.	The permission questions an agent asks you.
Connector (MCP)	A plug from an AI into an account, like email or calendar. MCP is the standard plug shape.	The connectors page in settings.
Temperature	The random dial for picking the next word. Low is steady. High is more varied and more risky.	Settings for people who build with AI. Why the same ask gives different answers.
Eval	A set of test questions with known right answers. You run it to check the AI still works.	Module 4. Job ads for people who build with AI.

How to use the sheet

Three habits make this more than a list.

When you read AI news, find which word it is really about. Most news is a bigger context window, a new tool, or a new model, dressed up.

When someone says "the AI got it wrong," ask which word fits. Maybe it was a hallucination. Maybe the context window filled up. Maybe a tool was missing. Maybe it needed news from after the cutoff. Each one has a different fix.

When a job ad says agentic, RAG, MCP, or evals, know that you have met all four. Modules 3 and 4 will have you use them. Dana got an email last month pitching "agentic, RAG-powered patient intake." Before this sheet, that would have meant nothing to her. After it, she knew to ask what data the agent could reach and what tools it could use.

Your task is to write the sheet in your own words. Do not copy the table. A meaning you can say from memory, your way, is one you own.