These are some of the terms you'll read and hear about. 45 terms, last reviewed 23-Sep-2026.
A
- Agentic AI
- AI that can plan and carry out multi-step tasks, using tools and taking actions to reach a goal, rather than just answering a single question. See also AI Agent.
- AI Act
- The EU's law regulating artificial intelligence, which sorts AI uses by risk. It can apply to UK businesses whose AI systems or output are used in the EU.
- AI Agent
- An AI system that is given a goal, access to tools (such as email, a calendar or a CRM) and rules about what it may do. It can then work through tasks with limited human intervention.
- API
- Application Programming Interface - a way for software to talk to other software. AI APIs let developers build AI into their own applications and are charged by usage.
B
- Batch API
- A way of sending lots of AI requests to be processed within a day instead of instantly, usually at around half the normal price.
C
- ChatGPT
- OpenAI's AI assistant, built on its GPT models. Available in free, paid and business versions (ChatGPT Business and Enterprise).
- Claude
- Anthropic's family of AI models and AI assistant, known for strong writing, careful reasoning, long documents and agent work. Current models include Claude Opus, Sonnet and Haiku.
- Context Window
- How much text an AI model can take into account at once, measured in tokens. A larger context window lets the AI work with longer documents or conversations.
- Copilot
- Microsoft's AI assistant. Microsoft 365 Copilot works inside Outlook, Word, Excel, PowerPoint and Teams using your own business data.
- Custom GPT / Project / Gem
- A saved, reusable AI assistant pre-loaded with your instructions and documents - called Custom GPTs or Projects in ChatGPT, Projects in Claude and Gems in Gemini.
D
- Deepfake
- AI-generated audio, images or video that convincingly imitates a real person. Increasingly used in fraud, such as fake phone calls from a "boss" asking for a payment.
E
- Embeddings
- A way of turning text (or images) into lists of numbers that capture meaning, so a computer can find content that is similar in meaning rather than just matching words. Used for searching your own documents.
F
- Fine-tuning
- Further training an existing AI model on a smaller, specific set of examples to make it better at a particular task. Less often needed now that models can simply be given your documents and instructions.
G
- Gemini
- Google's family of AI models and AI assistant, built into Google Workspace (Gmail, Docs, Sheets) and available through Google's API.
- Generative AI
- AI that creates new content - text, images, audio, video or code - rather than just analysing or classifying existing data.
- GPT
- Generative Pre-trained Transformer - the name of OpenAI's family of AI models behind ChatGPT.
- Grounding
- Connecting an AI's answers to reliable sources, such as your own documents or a web search, so that it answers from facts rather than memory. Reduces hallucinations.
- Guardrails
- Rules and checks built around an AI system to keep it on task, safe and within limits - for example stopping it discussing topics outside your business or taking certain actions.
H
- Hallucination
- When an AI states something that sounds convincing but is false. Always check important facts, figures and references in AI output.
- Human-in-the-loop
- A design where a person reviews or approves the AI's work or actions before they take effect - essential for important decisions.
I
- Inference
- Using a trained AI model to produce an answer. Each time you send a prompt and get a response, that is inference.
L
- Large Language Model (LLM)
- An AI model trained on vast amounts of text to understand and generate language. GPT, Claude, Gemini, Llama and Mistral are all LLMs.
- Local / On-premise AI
- Running an AI model on your own computer or server rather than in the cloud, so data never leaves your premises.
M
- Machine Learning
- The branch of AI where computers learn patterns from data rather than being given explicit rules. It is the foundation of almost all modern AI.
- MCP (Model Context Protocol)
- An open standard, introduced by Anthropic and now widely supported, for connecting AI assistants and agents to business tools and data sources in a consistent, controlled way.
- Model
- The trained AI "engine" that produces answers. Different models vary in capability, speed and price.
- Multimodal AI
- AI that can work with more than one type of input or output - text, images, audio and video. All the leading models are now multimodal.
N
- Neural Network
- A computing system loosely inspired by the brain, made of layers of connected "neurons". Deep neural networks are the basis of modern AI.
O
- Open-weight Model
- An AI model whose trained parameters are published so anyone can download and run it, for example Llama, Mistral, Gemma, gpt-oss and Qwen.
P
- Prompt
- The instruction or question you give an AI, including any documents or examples.
- Prompt Caching
- A saving offered by AI providers: when the same instructions or documents are sent repeatedly, the repeated part is charged at a much lower rate.
- Prompt Engineering
- The skill of writing clear, well-structured prompts to get better results from AI.
- Prompt Injection
- An attack where hidden instructions are placed in content an AI reads (an email, web page or document) to try to make it do something it shouldn't. A particular risk for AI agents.
R
- RAG (Retrieval-Augmented Generation)
- A technique where the AI first searches your own documents or data for relevant information, then uses it to answer. This is how most "chat with your documents" tools work.
- Reasoning Model
- An AI model that works through a problem step by step before answering. Better for maths, analysis and planning, but slower and more expensive.
- Rubber Duck
- An essential debugging tool for all programmers. Named after the practice of explaining your code line by line to a rubber duck in an attempt to find errors. The duck's ability to solve programming issues is legendary and not fully understood by science.
S
- Shadow AI
- Staff using AI tools the business hasn't approved, often on personal accounts - a common way for business data to leak.
- Small Language Model (SLM)
- A smaller, faster AI model designed for simpler tasks or for running on ordinary computers and phones.
T
- Temperature
- A setting that controls how varied an AI's responses are. Low values give consistent, focused answers; higher values give more creative, varied ones.
- Tokens
- The pieces of text an AI reads and writes - roughly three-quarters of a word each. AI API usage is priced per million tokens. For example, "Understanding how tokens work." is about 5 tokens: "Under", "standing", " how", " tokens" and " work."
- Training
- The process where an AI model learns patterns from large amounts of data. Business users rarely train models themselves - they use and configure existing ones.
- Transformer
- The type of neural network design behind modern language models - the "T" in GPT. It is particularly good at understanding context.
V
- Vector Database
- A database designed to store embeddings and quickly find the most similar items - used behind the scenes in RAG and document search tools.
- Vibe Coding
- Creating software by describing what you want to an AI in plain English and letting it write the code. Useful for prototypes; production software still needs proper review.
- Voice Agent
- An AI that holds spoken, real-time conversations - for example answering phone calls and booking appointments.