You’re in a meeting with an AI vendor, or a consultant, or a well-meaning employee who’s been doing their research. They start talking about agents and models and prompts, and somewhere around minute three, you stop tracking the content and start tracking how long until you can politely end the conversation.
That feeling isn’t a sign you’re not cut out for this. It’s a sign nobody took five minutes to give you the vocabulary. So here it is: five terms, explained plainly, that will unlock most of the conversations you’ll need to have.
Why these five terms specifically
In my work with founders and small business owners, there’s a consistent pattern in how AI conversations break down. It’s rarely about the technology being too complex. It’s that the person making the decisions doesn’t have a shared language with the people implementing it. Five terms come up again and again: prompt, model, hallucination, automation, and agent. Get comfortable with those five, and you can hold your own in any serious AI conversation without needing to understand anything about how the underlying systems actually work.
Think of it like buying a house. You don’t need to know how the electrical wiring works. But if you know what “load-bearing wall” means, you can have a much more useful conversation with your contractor.
The first three terms: what AI is and how it fails
Prompt is the simplest one. It’s just the instruction you give to an AI tool. When you type something into ChatGPT, that’s your prompt. When a business builds a custom AI tool that automatically summarizes customer emails, the instructions baked into that tool are prompts. The quality of what the AI produces depends heavily on the quality of the prompt. Bad instructions in, mediocre output out. This is why there’s an entire practice around “prompt design” — it’s just the craft of writing clear instructions for AI the same way you’d write a clear brief for a new hire.
Model is the AI itself — the thing doing the thinking. ChatGPT is an interface; GPT-4 is the model underneath it. Claude is a model made by a company called Anthropic. You’ll hear model names thrown around in vendor conversations. You don’t need to evaluate them technically. What you do need to know is that different models have different strengths, costs, and limitations — and the right model for one job isn’t always the right one for another. Your job is to ask: “Why this model for our use case?” Not to know the answer yourself.
Hallucination is the term for when an AI confidently states something that isn’t true. It’s not lying in any meaningful sense — it’s generating text that sounds correct without having any mechanism to verify whether it actually is. A client of mine discovered this when they asked an AI tool to pull competitor pricing from the web, and it returned figures that looked precise and professional and were entirely fabricated. The fix isn’t to stop using AI. It’s to understand where hallucination risk is high (research, facts, numbers) and build in human review at those points. Every serious AI implementation accounts for this.
The next two terms: what AI can do for you
Automation is the older of the two concepts, and the one you’re most likely already familiar with in some form. Automation means a system completes a task on its own once it’s been configured — no human hand required each time. Scheduling software that sends reminders. An email that goes out when a purchase is made. The difference with AI-powered automation is that the tasks don’t have to be simple or rigid. AI can handle variability. It can write a personalized follow-up email, not just send a templated one. That’s the meaningful expansion AI brings to automation.
Agent is the term that’s generating the most confusion right now — and the most hype. An AI agent is a system that can take multiple steps toward a goal, making decisions along the way, without a human directing each one. If automation is “send this email when X happens,” an agent is “here’s the goal, figure out the steps.” A simple example: an AI agent that notices a customer complaint, looks up the order history, drafts a resolution, and flags it for human approval before sending. It’s stringing together multiple actions toward a single outcome. Agents are powerful and genuinely useful. They’re also more complex to build well — which means any vendor pitching you an “AI agent” deserves more scrutiny, not less. Ask them to walk you through exactly what it does, step by step.
What to do with these terms this week
The next time someone pitches you an AI product or brings you an AI idea, use these five terms as a filter. Ask: what’s the model, and why did you choose it? How are we handling hallucination risk? Is this automation or an agent, and what’s the difference in what I’m paying for?
You’re not trying to stump anyone. You’re trying to see whether the person across the table has thought it through — and whether they can explain it to you clearly. The vendors and consultants worth working with will welcome those questions. The ones who can’t answer them plainly are telling you something important.
Vocabulary doesn’t make you technical. It makes you a better client, a better decision-maker, and a lot harder to impress with buzzwords alone.
