Agent
An AI that doesn't just answer - it takes action. It can book a meeting, reply to a customer or move a file across your tools. Think of it as a digital colleague you can hand a task to.
The words agencies and tech people throw around - explained the way a friend would. No jargon, just meaning.
An AI that doesn't just answer - it takes action. It can book a meeting, reply to a customer or move a file across your tools. Think of it as a digital colleague you can hand a task to.
An approach where the AI plans, tries, checks its work and adjusts - instead of just replying once. It's the difference between a chatbot and an autopilot.
A bridge that lets two pieces of software talk to each other. When your website sends a lead into your CRM, it's using an API. It's the glue of every modern integration.
Letting software handle a repeatable task for you - sending invoices, sorting leads, tagging emails. With AI in the mix, automations can now handle work that used to need a human judgement call.
A program that has a conversation with your users. The old ones followed scripts; the new ones use large language models and can actually understand what people mean.
OpenAI's chat assistant, built on their GPT models. The tool that put AI on everyone's desk. Good at writing, summarising, coding and thinking out loud.
Anthropic's AI assistant, a direct alternative to ChatGPT. Known for being thoughtful, careful with nuance and strong at long documents and writing. A favourite for teams that care about tone and safety.
How much text an AI can 'see' at once - your question plus any documents or history you give it. A bigger context window means the model can hold more of your project in mind at the same time.
Microsoft's AI assistant baked into Office, Windows and GitHub. Instead of a separate chat window, it sits inside the tools you already use - drafting emails in Outlook, formulas in Excel, code in your editor.
The branch of AI that uses layered networks loosely inspired by the brain. It's the engine behind image recognition, voice assistants and every large language model today.
The kind of AI that generates images by starting with noise and 'un-blurring' it into a picture. It's what powers tools like Midjourney, DALL·E and Stable Diffusion.
A way of turning text (or images, or audio) into a list of numbers that captures its meaning. It's how AI can tell that 'invoice' and 'bill' are talking about the same thing.
Taking a general AI model and training it a bit more on your own examples so it sounds like you and knows your specifics. Useful when prompts alone aren't enough.
Google's family of AI models and its ChatGPT competitor. Tightly integrated with Google Workspace, so it can pull from your Gmail, Docs and Calendar when you let it.
Short for Generative Pre-trained Transformer - the family of models behind ChatGPT and many others. In plain terms: a very good next-word predictor that turned out to be able to reason.
The rules and checks you wrap around an AI so it stays on-topic, on-brand and out of trouble. Things like 'never quote a price', 'always hand off to a human for refunds', or 'refuse legal advice'.
When an AI confidently makes something up - a fake source, a wrong number, a person who doesn't exist. It's the single biggest thing to guard against when you put AI in front of customers.
A setup where the AI proposes and a person approves before anything ships. The safest way to roll AI into work that touches customers, money or legal.
The moment an AI actually answers you. Training is when the model learns; inference is when it works. Every ChatGPT reply is an inference.
The set of your own documents, articles or data that an AI is allowed to draw from. Combined with RAG, it's how you make an assistant that actually knows your business.
How long you wait between asking the AI something and getting an answer. Low latency feels like a conversation; high latency feels like a form submission.
Large Language Model - the kind of AI that powers ChatGPT, Claude and Gemini. Trained on huge amounts of text, so it can write, summarise, translate and reason about almost anything in language.
Model Context Protocol - a shared standard that lets AI assistants plug into your tools (Google Drive, GitHub, your database) in a safe, predictable way. Think of it as USB for AI.
A popular image-generation tool known for its cinematic, art-directed look. You describe a scene in words and it paints it - great for moodboards, concepts and marketing visuals.
An AI that can handle more than just text - images, audio, video, spreadsheets. You can show it a screenshot and ask what's wrong; it can read it.
An AI model whose weights are published so anyone can download, run and modify it. Meta's Llama and Mistral's models are the best-known examples. Useful when you want full control or need to run AI on your own servers.
The company behind ChatGPT and the GPT models. One of the two or three labs setting the pace for the whole industry.
An AI-powered search engine that answers your question directly, with sources you can click. A common replacement for Googling when you want a summary, not ten blue links.
What you type to the AI. The clearer the prompt, the better the answer. Half of 'using AI well' is really 'writing good prompts'.
The craft of writing prompts that get you consistent, useful answers. Less mystical than it sounds - mostly it's being specific, giving examples and telling the model what to skip.
Retrieval-Augmented Generation. Instead of hoping the AI remembers a fact, you give it your own documents to read first, then it answers. It's how you make an AI trained on your business, not the internet.
A newer type of AI that 'thinks' before it answers - working through the problem step by step. Slower, but much better at maths, code and multi-step decisions.
OpenAI's video-generation model. You describe a scene and it produces a short video clip. Early days, but a preview of where marketing and product video are going.
The little chunks of text an AI reads and writes - roughly ¾ of a word. Pricing and context limits are usually measured in tokens, not words.
The long, expensive process of teaching an AI model by showing it huge amounts of data. Done once by the model makers; you rarely need to do it yourself.
A database built to store embeddings and find similar ones fast. It's the memory layer behind most 'chat with your documents' features.
Building software by describing what you want in plain language and letting an AI write the code. You steer, review and refine - the AI handles the syntax. It's how non-developers ship real products today.
A chain of steps - often across several tools - that runs automatically when something happens. New lead comes in → enrich it → score it → notify the right person. AI slots into any of those steps.
If you keep seeing a term and can't find a straight answer for it, send it our way. We'll add it in plain English.
Suggest a term →