Plain-English AI glossary
Short, jargon-free definitions of the AI terms businesses actually run into — each with a real-world example. The kind of clear answer you’d want whether you asked one of our Certified Claude Experts or asked your AI assistant. No hype, no fabricated numbers.
Agentic AI
The adjective for AI that behaves like an agent — taking initiative and working through steps toward a goal, rather than only answering the question in front of it. It’s the same idea as an AI agent, just described as a quality: if a tool is “agentic”, it can act like an agent. Don’t let the jargon make it sound like a separate technology.
Example A tool described as “agentic” could be handed the goal “sort these enquiries” and figure out the steps itself, instead of waiting for each instruction.
AI agent
An AI system you give a goal to, which then works out and carries out the steps to reach it — deciding what to do next, using tools or software along the way, and adjusting as it goes — instead of just answering a single question. A chatbot answers; an automation follows fixed steps; an agent is trusted to find its own path.
Example Given “draft a reply to this after-hours enquiry”, an agent could read the message, check your service-area list, look up the standard call-out guidance, and prepare a draft for a staff member to approve.
AI automation
Getting software to carry out a task by following fixed rules you set in advance — “when this happens, do that” — the same way every time. There’s no interpretation involved, which makes it reliable and easy to trust. Best for repetitive, predictable work with one correct outcome.
Example Every time a booking email lands, the customer’s name, job and date are pulled out and dropped into a spreadsheet and the calendar — automatically, every time.
AI workflow
A set sequence of steps where AI makes a judgement call at one or more points — reading, drafting or deciding based on what’s in front of it — rather than following a fixed rule at every step. Often a person checks or approves the key steps, which lets a workflow handle tasks that don’t look the same twice. Best for work that varies.
Example A client emails a question; the workflow drafts a tailored reply and hands it to a staff member to review and send.
Claude skill
A reusable package of instructions — and, if useful, example files — that teaches Claude how to do a specific task your way, so you don’t have to re-explain it each time. Claude brings the skill in when it’s relevant. Different from a one-off prompt: a skill is saved, shareable and reused.
Example A firm builds a “client onboarding email” skill — house tone, document checklist, structure — so anyone can produce a consistent draft for a new client.
Context window
The amount of text an AI can hold in mind at once — your conversation, anything you’ve pasted in, and its own replies so far. Go past it and the earliest material drops off, and the AI can forget what was said or shared at the start. Bigger models can hold more, but every one has a limit.
Example Paste one long contract and the AI can discuss all of it; paste a dozen and it may lose track of the first before you get to the last.
Fine-tuning
Further training of an existing AI model on extra examples so it gets better at a specific style or task. It changes the model itself — unlike a prompt or a skill, which guide the model without altering it. It’s usually a job for technical teams, and most businesses get what they need from good prompting and skills long before fine-tuning is worth it.
Example A large team with a very particular writing style might fine-tune a model on past documents — but for most, a well-built skill does the job with far less effort.
Hallucination
When AI states something that sounds confident and plausible but is actually wrong or made up. It’s a known, common failure mode — the AI isn’t lying, it’s predicting likely-sounding text — and it’s the main reason you always check important output before relying on it. “Confident but wrong” is the thing to watch for.
Example Asked about a rule it’s unsure of, an AI might invent an official-sounding clause that doesn’t exist — stated with complete confidence.
LLM (large language model)
The kind of AI behind assistants like Claude and ChatGPT. It’s trained on huge amounts of text to predict likely words, and that ability to predict is what lets it write, summarise, answer questions and hold a conversation in natural language. If a chatbot is the app you talk to, the LLM is the engine under the bonnet.
Example When you ask an assistant to summarise a report, it’s an LLM doing the reading and writing behind the scenes.
MCP (Model Context Protocol)
An open standard, introduced by Anthropic, for connecting AI assistants to outside tools and data sources in a consistent way — your files, a calendar, a business system. Think of it as a common plug: instead of a custom connection for every tool, MCP gives them a shared way to talk, so AI can safely reach the things it needs.
Example Through an MCP connection, an assistant could look up the right document from your files before answering a staff question, rather than guessing.
Multimodal
AI that can work with more than just text — images, audio, and sometimes documents or video — not only read and write words. It means you can show the AI something, not just describe it.
Example Photographing a damaged part and asking the AI what it is and whether it’s the sort of thing you’d usually replace.
Orchestrator agent
An AI agent whose job is to coordinate other agents or steps — breaking a big goal into parts, handing each to the right place, and pulling the results back together. If several agents each handle a piece of a job, the orchestrator is the one keeping the whole thing on track. The project manager of a multi-agent setup.
Example For “prepare this month’s client report”, an orchestrator might send one agent to gather figures, another to draft the summary, then assemble the result.
Prompt
The instruction or question you give an AI. Everything you type to get a response is a prompt — and the clearer and more specific it is, the better the result. A vague prompt gets a vague answer; a prompt with context, examples and clear limits gets something you can actually use.
Example “Rewrite this quote email to be friendlier and under 120 words, keeping the price and the two dates” is a far better prompt than “fix this email”.
Prompt engineering
The skill of writing clear, well-structured prompts to get reliable results — giving the AI context, examples and limits instead of a vague one-liner. Despite the name, it’s less “engineering” and more learning to brief well: the same clarity you’d give a capable new staff member so they get it right the first time.
Example Adding “here are two past examples of the tone I want, and don’t mention pricing” to a request is prompt engineering in action.
RAG (retrieval-augmented generation)
A method where the AI looks up relevant information from your own documents or a database and uses it to answer, instead of relying only on what it learned in training. It’s how you get answers grounded in your business’s actual content — and it helps reduce hallucination, because the AI is working from real source material rather than memory.
Example A staff chatbot that answers “how much leave do I have to give?” by pulling the exact wording from your own policy handbook.
Token
The small chunks of text an AI reads and generates — roughly a word or a piece of a word. Models break text into tokens to process it, and usage and limits (including the context window and, on paid plans, cost) are often measured in tokens. As a rough guide, a token tends to be a bit shorter than an average English word.
Example A short email might be a few hundred tokens; a long report, many thousands — which is why very long documents can bump into limits.
Learn to actually use this, on your own tasks
Knowing the words is a start — our AI Training & Enablement program takes non-technical Australian teams from the vocabulary to working AI builds on the jobs they do every day.
Book a call →