You’re sitting across from someone who wants to sell you an AI tool, or you’re on a call with a consultant, or you’re reading a proposal. The words are familiar enough that you don’t want to stop and ask. But underneath the nodding, you’re tracking maybe 60% of what’s being said.
That gap is not a you problem. It’s a vocabulary problem. And it turns out you only need five words to close it.
## Why this particular five
Most AI conversations that go sideways do so in the first ten minutes, before anything technical has even been discussed. A vendor says “we can build you an agent” or “the model might hallucinate” and the business owner hears something that sounds either impressive or alarming, with no way to tell which.
These five terms come up in nearly every conversation I have with clients about AI adoption. Learn them once, in plain English, and you’ll be able to follow any pitch, read any proposal, and ask the questions that actually matter.
You do not need to memorize definitions. You need to get the shape of each idea well enough that you can work with it.
## The five terms, in order of when you’ll encounter them
**Prompt**
A prompt is the instruction you give to an AI tool. That’s it. When you type into ChatGPT, everything you type is a prompt.
The reason this word matters is that the quality of what you get back depends heavily on how you ask. Think of it like briefing a new employee. “Write me a proposal” gets you something generic. “Write me a one-page proposal for a boutique hotel client who cares about sustainability and has a budget of under $50,000” gets you something useful. Same tool, completely different result. Better prompts produce better outputs, and learning to write them is a real skill — one you can build without any technical knowledge.
**Model**
A model is the AI engine underneath a tool. When someone says “GPT-4” or “Claude” or “Gemini,” they’re talking about a model.
Here’s the analogy I use: a model is like an oven. Different ovens have different strengths — some run hotter, some are better for particular things, some cost more to run. The tool you use (say, ChatGPT) is the kitchen. The model is what’s doing the actual cooking. You don’t need to understand how the oven was built to decide whether it’s right for your recipes. But knowing that models differ — in capability, cost, and what they’re good at — helps you ask better questions when someone is recommending one to you.
**Hallucination**
This one sounds alarming. It isn’t, once you understand what it means.
A hallucination is when an AI produces something that sounds confident and correct but is factually wrong. It might cite a study that doesn’t exist, invent a statistic, or describe a process that nobody actually follows. This happens because AI models work by predicting plausible-sounding responses, not by looking things up in a verified database.
The business implication is simple: anything an AI produces that contains facts, figures, names, or citations needs a human to check it. AI is an excellent first drafter. It is not a reliable researcher. Once you know that, hallucinations stop being scary and start being manageable — you just build a review step into your process.
**Automation**
Automation means setting up a system to do a task repeatedly, without a person having to trigger it each time.
This is not new to AI — businesses have automated things for decades. What AI adds is the ability to automate tasks that used to require judgment. Sorting emails. Summarizing documents. Drafting routine responses. Before AI, automation was mostly for rules-based, predictable tasks. Now it extends to anything with a pattern.
If someone proposes “automating” something in your business, the right question is: what currently triggers this task, who currently does it, and what does a good output look like? Those three questions will tell you whether the automation being proposed is solid or speculative.
**Agent**
This is the word that causes the most confusion, and also the most inflated promises.
An agent is an AI that can take a sequence of actions on its own, not just respond to a single question. Instead of answering one prompt, an agent might research a topic, draft a document, check it against criteria, and send it somewhere — all without you touching it at each step.
Think of the difference between a temp worker who needs a new instruction for every task, and a trusted team member who can take a goal and figure out the steps themselves. Agents are closer to the second. They are also newer, more complex, and more prone to going off-script. When a vendor promises you an agent that will “run your marketing” or “handle your customer service,” they’re describing something real in concept but often oversold in practice. Useful agents tend to be narrow and well-defined, not broad and autonomous.
## What to do with these five words
The next time you’re in a conversation about AI — a sales call, a vendor demo, a conversation with someone on your team — listen for these five words. When you hear one, you now have enough of a frame to ask a grounding question.
If someone mentions a model: “What model are you using, and why that one over the alternatives?”
If someone mentions an agent: “What exactly are the steps it would take, and where does a human review what it’s done?”
If someone mentions automation: “What’s the trigger, and what does an error look like?”
Those questions will tell you more about whether someone knows what they’re doing than any amount of technical vocabulary will.
You don’t need to become fluent in AI. You need enough language to stay in the room — and these five words will get you there.
