RESOURCES
Glossary
Plain-English explanations of the AI terms business owners actually need to know — what they mean, why they matter, and what to do about them.
A
Agent
An AI that can take actions on its own — sending emails, booking meetings, running searches, executing tasks across multiple apps.
Why it matters: This is where AI is heading next. Instead of you asking the AI to draft an email, an agent could draft it, send it, follow up, and book the meeting. Powerful — and easy to set loose without a strategy.
What to do: Don’t deploy agents on important work until you’ve used the AI manually for the same task long enough to know what good looks like. You can’t supervise something you don’t understand.
C
Context Window
The amount of information an AI can hold in its working memory during a single conversation.
Why it matters: When you start a new chat, the AI forgets everything from the last one. That’s why the same question can give you different answers on different days — you’re not giving it the same context.
What to do: Write a short business brief — a paragraph about your company, your customers, your goals — and paste it at the start of important sessions. Same context in, more consistent answers back.
F
Fine-Tuning
Why it matters: It sounds appealing — “an AI trained on your business” — but it’s expensive, technical, and almost always overkill for owners. A good prompt and good context will get you 90% of the way there for 1% of the cost.
What to do: Don’t pay anyone to fine-tune a model for you until you’ve exhausted prompting and RAG. Most of the time you’re better off improving how you use the tool than rebuilding the tool.
G
GIGO (Garbage In, Garbage Out)
A reminder that the quality of what you put in determines the quality of what you get out.
Why it matters: When owners say “AI gave me a generic answer,” nine times out of ten they gave it a generic prompt. The AI didn’t fail. The input did.
What to do: Before blaming the tool, look at what you typed. Add context. Add specifics. Ask for the format you want. Then look again.
H
Hallucination
When an AI confidently makes up information that sounds true but isn’t.
Why it matters: This is the single biggest reason owners get burned. The AI will invent statistics, cite fake sources, and make up case studies — all in the same confident tone it uses for real answers.
What to do: Verify anything that goes outside your business. Names, numbers, quotes, laws, dates — check them. Use AI for thinking, not for facts you can’t verify.
I
Inference
What’s happening when you send a prompt and the AI generates a response — the actual thinking, in real time.
Why it matters: Inference is what you’re paying for and what you’re waiting on. Faster inference means faster replies. Cheaper inference means tools can do more for less.
What to do: Mostly nothing. But when you hear vendors talk about “inference costs dropping” or “faster inference,” now you know they’re talking about the cost and speed of the AI doing its job — and that those numbers keep getting better.
L
Large Language Model (LLM)
An AI system trained on enormous amounts of text that can understand and generate human-like writing.
Why it matters: ChatGPT, Claude, and Gemini are all LLMs. When you hear “AI” in a business context today, this is almost always what people are talking about.
What to do: Pick one and learn it well. Bouncing between three different LLMs every week means you never get good at any of them.
M
Multimodal
An AI that can work with more than just text — images, audio, video, and documents too.
Why it matters: You can hand the AI a screenshot of your dashboard, a photo of a whiteboard, or a contract PDF and have a conversation about it. That changes what’s possible by an order of magnitude.
What to do: Stop typing things you could just show. Screenshot the email thread. Photograph the whiteboard. Drop in the PDF. Then ask your question.
P
Prompt
What you type into the AI to get it to do something.
Why it matters: The quality of your prompt determines the quality of your answer. Vague in, vague out.
What to do: Stop typing one-line questions. Tell the AI who it’s talking to, what you’re working on, what you want, and what good looks like. Two extra sentences up front saves twenty minutes of back-and-forth.
Prompt Engineering
The practice of writing prompts that get better answers from AI.
Why it matters: There’s a real skill gap between owners who get gold out of AI and owners who get garbage. The difference is rarely the tool — it’s the prompt.
What to do: Don’t buy a $500 prompt course. Learn three habits: give context, be specific about the output you want, and ask follow-up questions instead of starting over.
R
RAG (Retrieval-Augmented Generation)
A method that lets AI pull information from your own documents, databases, or files when answering a question — not just what it learned in training.
Why it matters: This is how AI gets to know your business specifically — your policies, your products, your customer history. It’s the difference between generic advice and advice grounded in your reality.
What to do: You don’t need to build a RAG system yourself. Tools like Claude Projects, ChatGPT custom GPTs, and NotebookLM all do a version of this. Upload your key documents and ask questions against them.
S
System Prompt
A behind-the-scenes instruction that tells the AI how to behave for an entire conversation or app.
Why it matters: This is how custom GPTs, Claude Projects, and AI features inside other apps get their “personality” and rules. You can use them yourself to set up a thinking partner that already knows your business.
What to do: In Claude Projects or ChatGPT custom GPTs, write a system prompt with your business context, your voice, and your rules. Use it for everything in that area of work. Now you’re not starting from scratch every time.
T
Temperature
A setting that controls how predictable or creative the AI’s answers are.
Why it matters: Lower temperature gives you consistent, conservative answers. Higher temperature gives you more variety and creative leaps. Most consumer AI tools hide this setting, but some let you tweak it.
What to do: For factual work — summaries, data extraction, contract review — use lower. For brainstorming, taglines, and creative work, use higher. If your tool doesn’t expose temperature, just ask the AI to “give me three very different versions” to get the same effect.
Token
A chunk of text the AI reads and writes in — roughly three-quarters of a word on average.
Why it matters: Pricing, speed, and context limits are all measured in tokens. When a tool says “1 million token context window,” it means it can hold about 750,000 words at once.
What to do: Don’t worry about counting them. Just know the term so you understand what you’re paying for and why long documents sometimes get cut off.
Training Data
The text and information an AI learned from before you ever talked to it.
Why it matters: An AI only knows what it was trained on, up to a specific date. That’s why it might give you outdated info on tax law, software versions, or current events.
What to do: For anything time-sensitive, either give the AI the current information yourself or use a tool with web search built in. Don’t trust it to know what happened last week.
