← All free tools
Free reference · Glossary

AI words, in plain English.

The words people use in meetings about AI, explained in a sentence, with what each one means on an ordinary Monday. Nobody has to admit they didn't know.

For: anyone who has nodded through a meeting about AI.

These are general explanations, not tied to any one product, and tools change quickly. Last checked November 2026.

Find a word

The basics

Generative AI: AI that makes new things: text, images, summaries, code. ChatGPT, Copilot, Claude and Gemini are all examples.On a Monday: it's the thing that wrote the first draft of the email you're now checking.
Large language model (LLM): The engine inside most AI assistants. It has learnt patterns from a vast amount of text and predicts what words should come next, very well.On a Monday: it's very good at sounding right, which isn't quite the same as being right.
Model: One particular version of an AI engine. Companies release new models every few months, and each behaves a little differently.On a Monday: the reason the same request gives a different answer after an update.
AI assistant, or chatbot: The window you type into. It sits on top of a model and handles the conversation.On a Monday: NEIL, keen and very quick.
Prompt: What you type or say to the AI: the question, the instruction, and any background you give it.On a Monday: a good prompt reads like a good brief. A vague one gets you something confident and generic.
Context, or context window: Everything the AI can see at once while it answers: your messages, any files you've added, its instructions. There's a limit to how much fits.On a Monday: in a very long conversation, the early instructions can quietly stop counting.
Token: The small chunks of text an AI reads and writes, roughly three quarters of a word each. Usage limits and costs are often counted in tokens.On a Monday: why the bill, or the "you've reached your limit" message, depends on how much you paste in.
Training data: The text and images a model learnt from before it was released.On a Monday: it knows a lot about the world in general and nothing about your business unless you tell it.
Knowledge cut-off: The date its training data stops. Unless it can search the web, it won't know what happened after that.On a Monday: last year's prices, presented as this year's.
Multimodal: An AI that works with more than text: images, voice, documents, sometimes video.On a Monday: you can photograph the whiteboard and ask it to type up the actions.

Working with it

Custom instructions, or system prompt: Standing instructions the AI follows every time, so you don't repeat yourself: who you are, how to write, what never to do.On a Monday: the place your team's AI rules should live, not just in a document nobody opens.
Project, or workspace: A folder in the AI tool with its own instructions and reference files, so every conversation in it starts with the same background.On a Monday: handy, until the files in it are two versions out of date and nobody owns them.
Knowledge base: The documents you give an AI to work from: price lists, policies, past work, house style.On a Monday: it will trust whatever's in there, including the price list from 2023. See the Brain That Grew.
Retrieval, or RAG: Short for retrieval-augmented generation. The AI looks things up in your documents before answering, instead of relying on memory.On a Monday: better answers about your business, as long as the documents are right.
Fine-tuning: Training a model further on your own examples so it behaves more like you want. More expensive and technical than instructions, and rarely the first thing a small firm needs.On a Monday: usually what someone means when they say "we'll train it on our data", and usually not what they actually do.
Copilot: Microsoft's name for the AI built into Office and Windows. People also use it loosely for any AI that sits inside software you already have.On a Monday: the "help me write" button that appeared in Outlook without a meeting about it.
Connector, or integration: A link that lets an AI read from or act in another system, such as your email, calendar, files or CRM.On a Monday: very useful, and worth knowing exactly what it can see and change.
MCP (Model Context Protocol): An open standard for connecting AI assistants to other tools and data, so one connector can work with several AI products.On a Monday: the plumbing behind "it can now look in our shared drive".
Automation, or workflow: A chain of steps that runs by itself: when this arrives, do that, then send this.On a Monday: saves hours, until step three fails quietly and nobody notices for a fortnight.

Agents

AI agent: An AI that works towards a goal over several steps by itself, using tools: searching, reading files, filling in forms, sending drafts.On a Monday: less a tool you use, more a junior colleague you've delegated to. See the agent line-up.
Tool use: When an AI can do things as well as write: search the web, run a calculation, open a file, call another system.On a Monday: the difference between "here's how you'd book it" and "I've booked it".
Multi-agent system: Several agents, each with its own job, handing work to each other. Often one coordinating agent passes tasks out.On a Monday: a team, which means it needs the same things a team needs, including someone who asks awkward questions.
Orchestrator: The agent, or the person, that splits a job into parts, hands them out and puts the results back together.On a Monday: the foreman. If nobody has that job, everybody assumes somebody does.
Memory: What an AI keeps between conversations: your preferences, past decisions, ongoing work. Some tools have it, some don't, and it's often editable.On a Monday: worth reading now and again. It remembers what it was told, not necessarily what's still true.
Human in the loop: A person who reviews or approves AI work at set points before it goes anywhere.On a Monday: necessary, not sufficient. Don't put a human in the loop. Put them in the design.
Guardrails: Limits built around an AI so it can't do certain things: send without approval, spend money, touch certain data.On a Monday: the rules that work even when nobody remembers them.
Persona: The character and way of working you give an AI: how cautious, how blunt, what it questions, who it answers to.On a Monday: the part most firms skip, which is why so many agents behave like the same keen new starter. That's Fabz.

When it goes wrong

Hallucination: When an AI states something false with complete confidence: a made-up source, a wrong figure, a policy that doesn't exist.On a Monday: NEIL, confidently wrong and already writing the next one. See the Confident Guess.
Sycophancy: The tendency of AI to agree with you and praise your ideas, because it's built to be helpful and pleasing.On a Monday: it called your plan excellent. It calls most plans excellent. See the Agreeable Assistant.
Bias: When AI repeats unfairness that was in the data it learnt from, for example in who it suggests for a role or how it describes people.On a Monday: a good reason a person always checks anything to do with hiring, pay or customers' circumstances.
Automation bias: The human habit of trusting a computer's answer over our own judgement, especially when we're busy.On a Monday: "it's usually right", said about the thing nobody has checked since March.
Model drift, or update changes: When an AI starts behaving differently because the model underneath was updated, or the data it works from changed.On a Monday: Derek saying "it never used to do that", and being right.
Prompt injection: Hidden instructions inside a web page, email or document that try to make an AI do something its user didn't ask for.On a Monday: one reason agents that read your inbox shouldn't be able to send money or files without a person saying yes.

Data and safety

Shadow AI: AI tools people use for work that the business hasn't approved or doesn't know about, often on personal accounts.On a Monday: very common, rarely malicious, and the starting point of Before We Switch It On.
Personal versus work account: Business versions of AI tools usually come with different terms about how your data is stored and used than free personal ones. Check the terms for the tool you use.On a Monday: the same tool, two very different answers to "where did that customer list go?"
Data retention: How long an AI provider keeps what you type and upload, and whether it can be used to improve its models. Settings and terms vary by product and plan.On a Monday: worth one person finding out properly, once, and writing it down.
Personal data: Anything that identifies a living person: names, emails, phone numbers, addresses, notes about them. In the UK it's covered by data protection law (UK GDPR).On a Monday: the line most teams' first AI rule is about. Ask your data protection lead, not the AI.
AI policy: A written statement of how a business uses AI: what's allowed, what isn't, who decides and who checks.On a Monday: a policy written for the handbook gets ignored. One the team helped write gets used.