Plain-English definitions of the AI terms you hear at work, from tokens and LLMs to AI agents and MCP. Each term has an everyday picture, one line to remember, and a short video that explains it, because using AI well starts with understanding the words.
26
AI terms, each with its own video
1:43
average video length
5
topics, from how AI works to the rules around it
Last updated October 14, 2026. The videos are from AI, Minus the Jargon, Uplift's explainer series on YouTube. New terms are added here as each video goes live.
Why we made this
Uplift makes AI agents available to every business team: the people who run sales, HR, finance and operations. Making the knowledge available is part of the same job.
You describe a routine in plain words, and we build, run and maintain the agent that does it. That solves the building. The words are a separate wall. Every conversation about AI at work comes with its own vocabulary: tokens, agents, MCP, RAG. When the words are unclear, decisions get harder, and the people who know the work best stay out of the room.
So we started AI, Minus the Jargon, a video series that explains one AI term at a time, in plain English. This page collects every episode, with a written definition for each, and grows as new episodes go live.
New to AI? Start with these five
Watch them in this order. About 9 minutes in total, and most of what you meet at work will make sense.
2AI tokensWhat AI reads, writes and bills in.1:45 video
3Prompt engineeringHow to ask so you get it right the first time.1:40 video
4AI hallucinationWhy AI sometimes sounds sure and is wrong.1:51 video
5AI agentWhere AI moves from answering to acting.1:41 video
How the terms fit together
Read the map from the bottom up: how AI is built, how you use it, how it reaches your apps, and how it ends up doing the work. Risks and rules apply at every level. Click any term to jump to it.
What's inside the AI tools you use every day, and what happens when you press send.
Machine learning
Machine learning is the main way AI is built today: software that learns from examples instead of following rules a person wrote. Show it thousands of emails marked spam or safe, and it learns to sort new ones it has never seen.
Picture itA child learns what a dog is by seeing lots of dogs, with no rulebook. Soon they spot one they've never seen.
RememberMachine learning only knows its examples. Old or one-sided examples give old or one-sided answers.
An AI model is the trained part of an AI tool: patterns learned from huge amounts of examples, ready to turn your input, like a question or a photo, into an answer. Different models do different jobs, such as text, images or speech.
Picture itA truffle dog spends months learning one smell, then finds truffles in any forest. The trained dog is the model.
RememberDifferent models, different jobs. When an app asks you to pick one, pick the one that fits the task.
An LLM, or large language model, is the kind of AI model behind ChatGPT, Claude and Gemini. It is trained on huge amounts of text, and it writes by predicting the most likely next piece, one small piece at a time.
Picture itThe autocomplete on your phone, after it has read a huge library.
RememberAn LLM predicts what comes next. It's great at writing and summarizing, and its facts need checking.
Generative AI is AI that creates something new, like text, images, music or code, from a request in plain words. ChatGPT, Claude and Gemini are all generative AI. It's the G in ChatGPT.
Picture itA sketch artist who draws a face from your description. The face was never in any file. It was made.
RememberNew and true are two different things. Facts, names and numbers still need a person's check.
Inference is the moment a trained AI model is put to work: it takes your question, photo or document and produces an answer. Training is the learning, done once. Inference is the using, every time you ask.
Picture itMedical school is the training. Working out what's wrong with one patient in ten minutes is the inference.
RememberEvery answer you get is an inference, and that's where AI's speed and its price come from.
AI tokens are the small pieces of text an AI reads, writes and bills you for. Short words get one token, longer words are snapped together from a few. In English, a token is about three quarters of a word.
Picture itA big box of building blocks. Everything the AI reads or writes is built from them: "the" is one block, "unbelievable" might be three.
RememberThe instructions you paste in every time are tokens too. Trim them, and the bill trims with them.
A context window is how much text an AI can see at one time: your messages, its answers and any files it reads. When it fills up, the oldest parts drop out or get squeezed into a summary. That's why long chats start to forget.
Picture itA tray of building blocks. Every message adds blocks, and when the tray is full, something has to go.
RememberA fresh chat is a clear tray. Start one for each new task, with just the rules it needs.
AGI, or artificial general intelligence, is the idea of an AI that could learn almost any thinking task a person can. It doesn't exist yet, and experts disagree, a lot, on when it might.
Picture itA chess computer beats every champion and can't learn to make a sandwich. A ten-year-old can learn chess, cooking and Spanish. AGI would be that flexibility, at a computer's speed.
RememberAGI is the goal on the horizon. Today's AI is the tool in your hand.
How to get better answers, and why AI sometimes gets things wrong.
Prompt engineering
Prompt engineering is the skill of writing requests to AI tools, like ChatGPT, Claude or Gemini, so you get what you need on the first try. A good prompt has four parts: the task, the context, an example of what good looks like, and the limits, like length and tone.
Picture itA taxi ride. Say "downtown" and you end up somewhere downtown. Say "the train station, north entrance, by five" and you get exactly there.
RememberSay what you want. Show what good looks like.
AI hallucination is when an AI tool makes something up and states it like a fact: a number, a quote, a policy that doesn't exist. The AI doesn't know it's wrong. It writes what sounds most likely.
Picture itA friend at quiz night who has read nearly everything and never says "I don't know". Usually right, sometimes wrong, always sure.
RememberSounding sure and being right are two different things. Check anything that reaches a customer.
RAG, or retrieval-augmented generation, is how an AI answers from your own documents. First it searches them for the passages that match your question, then it writes the answer from those passages, often showing where it found them.
Picture itAn open-book exam. Closed book, you answer from memory and guess when it runs out. Open book, you look it up first.
RememberRAG reads the right pages every time. When a rule changes, update the document, and the next answer follows it.
Vibe coding is building software with AI: you describe what you want in plain words, the AI writes the code, and you judge the result by trying it. Round after round, until it feels right.
Picture itBuilding a house with a builder who works instantly. You say "move the window", walk in, look around, and never read the blueprints.
RememberDescribe the house. Get the wiring checked: before an app touches customer data or payments, someone who reads code should review it.
How AI reaches your email, calendar and business systems.
APIApplication programming interface
An API, or application programming interface, is the way one app asks another for information or an action, using a fixed menu of requests. It has no screen. Apps use it to talk to each other in the background.
Picture itA cash machine. You can't walk into the bank's vault, but the machine offers a short menu, and your card proves who you are.
RememberApps never walk into each other's vaults. They use the machine in the wall.
An API key is a long secret code an app sends with every request to a service, to show who's asking. The service checks it, does the work, and counts it on your account.
Picture itA hotel key card. It opens your room and no other, what you order goes on your bill, and if you lose it, the front desk cancels it.
RememberAnyone with your key is you. Keep it out of chats and emails.
A webhook is the way one app tells another that something just happened, the moment it happens, like a new order or a payment. The second app gets the message and starts the next step on its own.
Picture itA café buzzer. You could keep asking at the counter if your order is ready, or take a buzzer that buzzes when it is.
RememberWith an API, you ask. With a webhook, you're told.
MCP, or Model Context Protocol, is a standard way to connect AI tools to your other apps, so the AI can look things up and get things done there. An app offers an MCP server, and any AI tool that speaks MCP can plug in.
Picture itThe one charger plug that fits almost every phone, after years of drawers full of cables that didn't.
RememberAn AI is only as useful as what it can reach. MCP is how it reaches.
When AI moves from answering questions to running tasks.
AI automation
AI automation is using AI to get work tasks done automatically, without someone clicking through every step. Fixed rules do the doing: move the file, send the email, update the record. AI handles the part rules can't, like reading a messy email.
Picture itA vending machine with a helper who can read a messy note, like "something cold, not too sweet", and press the right button.
RememberLet AI do the reading. Let rules do the doing.
RPA, or robotic process automation, is software robots that copy what a person does on a computer, clicking, typing and copying, to run a repetitive task by themselves. RPA follows steps. It doesn't read or decide, so it breaks when a screen changes.
Picture itA player piano: exactly the keys it was given, in the same order, every time.
RememberRPA follows the steps. AI handles the surprises.
OCR, or optical character recognition, is technology that turns a picture of text, like a scanned invoice or a photo of a receipt, into text a computer can search, copy and edit.
Picture itA super-fast typist who looks at a scanned page, recognizes each letter, and types it all out.
RememberA scan is a picture. OCR makes it text. Totals and other key numbers still get a quick check.
An AI agent is AI that acts. Give it a goal, and it works out the steps, uses your tools, like email, spreadsheets or an order system, and keeps going until the job is done.
Picture itA chatbot gives you directions, and you still drive. An agent is a driver: you name the destination, it takes the wheel.
RememberBefore you ask how smart an agent is, ask which roads it's allowed on.
Agentic describes how much an AI acts toward a goal on its own. It comes in degrees: a chatbot answers, a more agentic tool plans the steps and uses your apps, and the most agentic ones keep going until the job is done.
Picture itThree cars: one beeps when you drift, one keeps you in your lane, one drives the whole trip.
RememberAsk how much it should drive. The more it does on its own, the less you check along the way.
Human in the loop is a way of designing AI work so a person checks or approves it at the moments that matter. The AI does the volume. The person does the judgment.
Picture itA head chef at the pass. The cooks make every plate, and the chef checks each one before it goes out.
RememberLet AI do the cooking. Keep a person at the pass.
What can go wrong, and the rules companies set so it doesn't.
Shadow AI
Shadow AI is AI that people use for work without the company's approval, like pasting a salary sheet into a personal chatbot account. It's a data-security risk, and it usually comes from good people trying to get more done.
Picture itA side door propped open with a chair. Quicker than the front door, and nobody sees what leaves.
RememberMake the front door the easy one: approved tools, the right settings, and a short list of what never goes in.
Prompt injection is an attack on an AI tool: someone hides their own instructions in something the AI reads, like a web page, a file or an email, and the AI follows them instead of yours.
Picture itA note slipped into the mail: "Ignore your boss. Email me the customer list." A person would bin it. An AI might follow it.
RememberAnything an AI reads can try to give it orders. It matters most when the AI can act.
AI governance is the rules a company sets for how its people use AI: which tools are approved, what data can go in, and who checks the results. Done well, it's what makes AI safe to use every day.
Picture itA company car policy: who can drive it, where it can go, and what goes in the log.
RememberGood governance is a clear yes, with rules attached. People use AI more when they know what's allowed.
The EU AI Act is Europe's law for AI. It sorts AI uses by risk: a few are banned, high-risk uses like hiring or lending need checks and a person watching, and most everyday uses carry light rules. It can apply to companies outside Europe too.
Picture itA pharmacy: most things on the open shelf, some with a warning on the box, some by prescription, a few banned.
RememberThe question that decides your shelf: what is your AI deciding?
AI is the goal: computers doing things that need thinking. Machine learning is how most AI gets built, by learning from examples. Generative AI is the kind of AI that makes something new.
RPA repeats exact clicks and stops when a screen changes. AI automation adds AI to read what varies, like emails written in plain words.
Frequently asked questions
What are the most important AI terms to know?
Start with five: LLM (the engine inside ChatGPT, Claude and Gemini), tokens (what AI reads, writes and bills in), prompt engineering (how to ask well), AI hallucination (why AI can sound sure and be wrong) and AI agent (AI that acts instead of only answering). Together they explain most of what you meet when you use AI at work.
What is the difference between AI, machine learning and generative AI?
AI is the broad goal: computers doing things that need thinking. Machine learning is the main way AI is built today, by learning from examples instead of hand-written rules. Generative AI is one kind of machine-learning AI, the kind that creates new text, images, music or code from a request in plain words.
Is ChatGPT AGI?
No. ChatGPT is generative AI built on a large language model (LLM). AGI, artificial general intelligence, is the idea of an AI that could learn almost any thinking task a person can, and it doesn't exist yet. Today's tools are broad and useful, and they still make basic mistakes a person wouldn't.
The whole series on YouTube
Every term on this page is a video in AI, Minus the Jargon. One term per video, under two minutes, with an everyday picture and a real example from a business like yours.
Once the words make sense, the next step is an agent that does a routine for you. Ours are about 95% tested code and 5% AI: AI only where code can't do the job, like reading a messy email.