AI, in plain English.

Large language model

You already know it asThe autocomplete on your phone, grown up and very well read.

The engine behind the chat assistants you already use. It has read a huge amount of text and predicts the next words well enough to write, summarise, translate and answer questions.

In a real business: You rarely pick the model yourself. You pick the tool built on it, and that choice matters more.

You already know it asA brief you give a freelancer.

The instruction you type to an AI. Vague brief, vague work. Say who it's for, what good looks like, and give an example.

In a real business: "Write a follow-up to a lead who went quiet after a quote, friendly, under 80 words" beats "write a follow-up email".

Also called custom instructions

You already know it asThe job description and house rules a new hire gets on day one.

Standing instructions an AI follows in every conversation: your tone, your offers, what it must never say.

In a real business: Your enquiry assistant always quotes your real prices and never promises a delivery date.

You already know it asThe size of the desk someone is working at.

How much an AI can keep in view at once: your instructions, the documents you gave it and the conversation so far. Go past it and the oldest parts fall off the desk.

In a real business: Why a long chat starts "forgetting" what you said at the start. Start a fresh one with a short summary.

You already know it asCall minutes or SMS units.

The small chunks of words an AI reads and writes. Paid AI tools charge by the token, the way phone plans charge by the minute.

In a real business: Only matters when you pay per use, for example an AI that answers hundreds of customer messages a day.

You already know it asA confident new hire who bluffs instead of saying "I don't know".

When an AI states something false as if it were fact: a made-up price, statistic, quote or policy.

In a real business: Never let AI send prices, legal wording or figures to a customer without a check, or without it reading them from your own documents.

You already know it asA colleague who can read, look at photos and listen to voice notes.

AI that works with more than text: images, screenshots, PDFs, audio and sometimes video.

In a real business: Snap a receipt or a delivery note and have it read, filed and logged for you.

You already know it as"Let me think about it and get back to you."

An AI model that works through a problem in steps before it answers. Slower and pricier, but better at planning, maths and tricky decisions.

In a real business: Worth using for a pricing model or a messy schedule. Overkill for a quick email.

Application programming interface

You already know it asThe order window at a restaurant.

The agreed way two pieces of software ask each other for things. You never see it, but every tool that "connects" to another uses one.

In a real business: Before you buy a tool, ask "does it have an API?" If not, joining it to the rest of your business will be painful.

You already know it asThe bank alert you get the moment money lands.

An automatic message one app sends another the instant something happens, so the next step can start without anyone checking.

In a real business: A payment clears, and the welcome email, invoice and onboarding call invite go out on their own.

Also called a workflow

You already know it asA standard operating procedure that runs itself.

A fixed set of steps that runs the same way every time something happens. No judgement, just reliable.

In a real business: New enquiry in, details saved, reply sent, you notified on WhatsApp. Start here before you think about agents.

Retrieval-augmented generation

You already know it asAn open-book exam, using your staff handbook.

An AI that looks things up in your own documents before it answers, so the answer comes from your information, not its general knowledge.

In a real business: An assistant that answers customer questions from your FAQs, policies and price list, in your words.

Also vector database

You already know it asA library catalogue sorted by meaning, not by title.

A way of storing text so a computer can find things that mean the same, even when the words differ. "Money back" finds your refund policy.

In a real business: It sits under most "chat with your documents" tools. Your builder chooses it; you don't need to.

You already know it asSending a staff member on a specialist course.

Training a model further on your own examples so a style or task becomes second nature to it.

In a real business: Rarely the first step. Good instructions plus your own documents (RAG) get most owners there faster and cheaper.

You already know it asA trusted assistant with a to-do list and the keys to your tools.

AI that can plan steps and take actions, not just chat: reply to a lead, check your calendar, update a record, then report back.

In a real business: An agent that answers quote requests at 2am, asks the right questions and books the call. Start with one narrow job.

Also called a copilot

You already know it asA capable intern at the next desk. You stay in charge.

AI built into a tool you already use that drafts, summarises and suggests, but waits for you to approve.

In a real business: Drafting proposals, summarising long email threads, turning meeting notes into a task list.

Also called function calling

You already know it asGiving a new hire logins and a company card, with limits.

The specific actions an AI is allowed to take, such as "check the calendar" or "create an invoice". No tools, no action.

In a real business: Decide which tools an agent gets the way you'd decide what a new hire can access in week one.

Model Context Protocol

You already know it asA universal plug adapter.

An open standard that lets AI assistants connect to your apps (email, calendar, files, CRM) the same way every time, instead of a custom connection for each.

In a real business: When choosing software, "does it work with MCP?" now tells you how easily AI can work with it.

You already know it asThe office an employee works in: desk, tools, rules and a manager.

The software wrapped around an AI model that gives it instructions, tools and memory, and checks its work. Same model, different harness, very different results.

In a real business: Why two products using the same AI can perform completely differently. Judge the setup, not the brand of model.

You already know it asA playbook or SOP binder handed to an assistant.

Packaged instructions and files an AI loads when a certain kind of task comes up, so it does that job your way every time.

In a real business: "How we write proposals" or "how we onboard a client", written once and followed every time.

Also multi-agent

You already know it asA manager handing parts of a job to specialists.

One AI splitting a big task and handing pieces to other AIs, each with its own focus, then pulling the results together.

In a real business: Useful for large research or build jobs. Most small businesses need one good agent first.

Also browser agents

You already know it asA virtual assistant working on your computer remotely.

AI that clicks, types and moves through screens like a person. Handy for old systems with no API.

In a real business: Slower and less reliable than a proper connection. Use it when there's no other way in.

You already know it asA receptionist, or the old "press 1 for sales" phone menu, that actually talks.

AI that answers and makes phone calls in natural speech: takes messages, answers questions, books appointments.

In a real business: Worth it if you miss calls. Every missed call is a lead someone else answers.

You already know it asSpending limits and approval rules.

Rules that stop an AI doing or saying what it shouldn't: topics it avoids, actions that need a human, limits it can't pass.

In a real business: The AI can offer a discount up to 10%. Anything more goes to you.

You already know it asManager sign-off.

A person checks or approves the AI's work at chosen points before it goes out.

In a real business: AI drafts every reply. You approve anything involving money, complaints or new prices.

Short for evaluations

You already know it asA probation review, or quality checks on a sample of work.

Tests that measure whether an AI does the job right, run before launch and after every change.

In a real business: Twenty real customer questions with the right answers, checked every time the assistant is updated.

You already know it asA phishing email, aimed at your AI instead of you.

Hidden instructions inside an email, web page or document that try to trick an AI into doing something you never asked.

In a real business: Matters the moment your AI reads outside content. Keep risky actions behind human approval.

You already know it asDescribing the house you want instead of drawing the plans.

Building software by describing it to AI in plain language and letting it write the code.

In a real business: Great for a quick prototype. Anything that takes payments or holds customer data still needs a proper review.