The words you'll hear in vendor calls, board meetings, and conference panels — defined in plain language by the people building this stuff for a living. Nineteen terms. No marketing copy. Share it with anyone who needs to follow along.
The words you'll hear in every AI conversation. If someone is using these without ever explaining them, ask. A good vendor will be happy to.
Artificial Intelligence
A computer program that can do things we used to think only people could do — read, write, summarize, answer questions, find patterns in messy information. Most of what people call "AI" today is one specific kind: a large language model.
When someone says "AI," ask which kind. There's a big difference between a chatbot, a tool that reads documents for you, and a system that takes real actions on your behalf.
Large Language Model
The thing behind ChatGPT, Claude, and similar tools. A computer program that learned how language works by reading a huge amount of text, and now uses that to answer questions, write drafts, and summarize information.
An LLM is good at language work — writing, summarizing, translating, finding the right paragraph in a 50-page document. It is not a database, and it does not always know what's true.
The AI itself, separate from the app you talk to. ChatGPT is an app; the model underneath it is called GPT-4 (or whichever version). Claude is an app; the model is also called Claude. The model is the engine, the app is the dashboard.
Different models have different strengths and costs. A good tech partner will pick the right one for the job, not just use the most expensive one for everything.
What you type to an AI. The instructions, the question, the document you paste in — all of it together is the prompt.
Most of the difference between a useful AI tool and a frustrating one is how the prompt is written underneath. That's part of the work we do.
An AI that can take actions, not just chat. A regular chatbot writes you a reply. An agent can read your email, draft a response, schedule a meeting, and send the message — if you tell it to.
Agents are powerful and risky in the same breath. You want them doing the boring repeat work, with a person checking the important moments. Never set an agent loose on something where a mistake costs money or hurts a relationship.
We use agents for the boring middle — pulling data, drafting summaries, formatting reports. We do not use agents for anything where a person should be making the call.
Application Programming Interface
How one piece of software talks to another. When your website checks the weather and shows it on the homepage, it's using the weather service's API. When an AI tool reads from your member database, same idea.
If a vendor says "our tool has an API," they mean other software can plug into it. That's usually good news — it means you're not stuck and you can connect things later.
The words that get thrown around in sales calls. Most of them mean something real. A few are dressed-up versions of plain ideas.
When an AI confidently makes something up. Cites a court case that doesn't exist. Quotes a study that was never written. Tells you a person works somewhere they don't.
Never trust an AI's first answer on anything that matters — names, dates, numbers, citations. The safe pattern is: AI drafts, a person checks before it goes out.
The text the AI learned from. Books, websites, articles, public documents. For the well-known AI tools, training data is everything on the internet up to a certain date.
An AI doesn't automatically know your specific information unless you give it to it. "It learned from the whole internet" doesn't mean it knows your member list.
How much an AI can hold in its head at one time. Modern AI can read a small book in one sitting. Older AI could only hold a few pages at a time.
When someone tells you AI can "read your whole knowledge base," ask how. Either it's all small enough to fit, or there's a search step that finds the right pieces first.
Retrieval-Augmented Generation
A fancy name for a simple idea: before the AI answers, it looks something up. Instead of relying only on what it learned, it pulls real information from your documents, then writes the answer.
This is how an AI tool can know your specific information — your policies, your member history, your past reports — without ever being "trained" on it. Your information stays on your side; the AI just looks it up when asked.
Taking an AI and putting it through extra schooling on your specific writing or data, so it sounds more like you. Different from giving it documents to read — fine-tuning actually changes how the AI itself works.
Most organizations don't need this. Good prompts plus the look-it-up approach (RAG) get you 90% of the way there for a fraction of the cost. We've never had a client who genuinely needed fine-tuning.
Getting stuck with one company because moving away from them would be too painful — your data is in their format, your team is trained on their tool, and starting over feels impossible.
The way to avoid lock-in is to own your information. If your member list lives in a system you own and the AI tools just look at it, you can swap AI tools whenever a better one comes along.
A part-time, senior tech person. Same level of experience as someone running technology at a big company, but renting a slice of their week instead of hiring them full-time. Sometimes called a "part-time CTO" or "tech advisor."
Most organizations under 50 staff don't need a full-time tech leader. They need someone senior to lean on a few hours a week — for decisions, vendor calls, and the question "is this a good idea?"
These are the words we use a lot. They're not industry-standard — they're shorthand for the way we approach this work.
The piece between "we have AI" and "AI actually helps our staff." Most organizations can buy or sign up for AI in an afternoon. The hard part is connecting it to the way your team works — which documents it reads, when it asks a person, where the output goes, who can use it.
When a vendor sells you AI, they're selling the engine. The implementation layer is the rest of the car — the steering wheel, the seats, the dashboard. That's where most of the real work happens, and it's what we do.
The repetitive, low-judgment work that happens between someone making a decision and someone receiving the result. Formatting a weekly report. Copying numbers between systems. Drafting the same email for the fortieth time.
This is where AI shines. It's also where your staff is losing hours every week. The first AI moves for most organizations are in the boring middle — not the fancy stuff.
We do not try to automate the parts of your work that require judgment, relationships, or cultural knowledge. The boring middle is the right size of bite.
A workflow where AI does most of the work, but a person checks the moments that matter before anything goes out. The AI drafts the email; a person reads it before send. The AI pulls the numbers; a person eyeballs them before they hit the board report.
This is the safe default. We design every workflow this way unless there's a strong reason not to. A two-minute review by a person catches mistakes before they become problems.
A simple rule: when we build something, your information lives in a place you own, not on someone else's servers. The AI is allowed to read it when asked. It is never used to train someone else's AI.
For tribal organizations especially, this matters. Your member information, your cultural knowledge, your enterprise data — none of it should be feeding a system you don't control.
We will never recommend a vendor that uses your information to train their AI. We will tell you when one tries to slip that into a contract.
Our shorthand for what good AI looks like at this stage. A weekly summary that pulls from the systems you already have, written in your voice, delivered to the right inboxes — for about two cents per send.
If a vendor is asking for $5,000 a month for something that should cost two cents to run, ask hard questions. The AI itself is cheap. What costs money is the work of setting it up correctly.
A printable PDF version of this whole glossary — designed to share with a board, a director, or a colleague who keeps getting AI words thrown at them. Drop your email; we'll send it. You'll join the occasional notes list — unsubscribe in one click.
Book a free 15-minute call. We'll define whatever's confusing in your specific situation, no slides, no pitch.
Book a free intro call