AI and automation glossary: 40 terms in one sentence each

The short answer

AI and automation have produced more vocabulary in a few years than most business technologies produce in decades, and much of it is used loosely in pitches and headlines. The forty terms below are the ones that actually come up in proposals, project updates and vendor conversations, each explained in one plain sentence. They fall into five groups: what the systems are, how they work, how they are connected, how they are kept safe, and how they are run and measured. When a term is used to sell, ask what it does for your process and what would happen without it. The answer is often something ordinary described grandly, and knowing that is the beginning of buying well.

What the systems are

Term	In one sentence
Automation	Software that performs a repeatable task by following rules, without judgement, reliably and cheaply.
Artificial intelligence	Software that performs tasks that would otherwise need human judgement, such as understanding language or recognising patterns, with some uncertainty.
Machine learning	A way of building AI by learning patterns from examples rather than by writing rules.
Large language model	A model trained on vast amounts of text that can read, write, summarise, classify and converse, and that is behind the current generation of AI tools.
Generative AI	AI that produces new content, text, images, code or audio, rather than only classifying or predicting.
Chatbot	A conversational interface, which may be rule-based, AI-based or both, and whose quality depends entirely on what sits behind it.
AI agent	A system that uses a model to plan and carry out multi-step tasks using tools, which is powerful and needs tight boundaries.
Classifier	A model or rule set that sorts inputs into categories, such as enquiries by type, and one of the most reliable uses of AI.
Extraction	Pulling structured fields, such as a total and a date, out of unstructured input such as an email or a scan.
Recommendation system	Software that suggests items or actions based on patterns in behaviour and data.
Forecasting model	A model that estimates future values, such as demand, as a range with a confidence, from historical data.
No-code and low-code	Tools for building automations with little or no programming, right for simple flows and limiting for complex ones.

How they work

Term	In one sentence
Prompt	The instruction and context given to a language model, which shapes what it produces more than most people expect.
Context window	The amount of text a model can consider at once, which limits how much of your data it can read in one request.
Token	The unit in which models read and write text and in which providers bill, roughly three-quarters of a word in English.
Grounding	Giving a model your own data and instructing it to answer only from that, so it does not invent.
Retrieval-augmented generation	Finding the relevant passages from your documents and giving them to the model to answer from, the standard way to use AI with company knowledge.
Fine-tuning	Further training a model on your examples to change its behaviour, rarely needed by small businesses compared with prompting and grounding.
Hallucination	A model producing confident, plausible content that is false, which grounding, guardrails and review exist to catch.
Confidence score	A model’s estimate of how sure it is, used to decide what flows automatically and what a person reviews.
Embedding	A numerical representation of text that lets systems find passages by meaning rather than by keyword.
Open model	A model whose weights are published, so it can be run on your own infrastructure or by a provider of your choice.

How they are connected

Term	In one sentence
Integration	A connection between systems so data flows without retyping, where most automation value lives.
API	An agreed way for one piece of software to ask another for something, the basis of every integration.
Webhook	A message one system sends another the moment something happens, so the second does not have to keep asking.
Workflow	A defined sequence of steps, triggered by an event, with rules for what happens at each step.
Trigger	The event that starts a workflow, such as a form submission or a new order.
System of record	The one system that holds the authoritative version of a kind of data, which every automation reads from and writes to.
Idempotent	A property of a step that produces the same result if run twice, essential when systems retry.

How they are kept safe and run

Term	In one sentence
Human in the loop	A person who reviews, approves or decides at defined points, so automation assists rather than replaces judgement.
Guardrails	Rules and checks around an AI system that limit what it may say, do, access or decide.
Prompt injection	An attack in which malicious instructions hidden in content trick a model into ignoring its rules, a reason to limit what models can do.
Data processing agreement	The contract under which a provider processes personal data on your behalf, required before customer data goes to any AI service.
Data minimisation	Sending a model only the fields a task needs, which reduces risk and often improves results.
Evaluation	Testing an AI system against a set of real examples with known good answers, before and after every change, to know that it works.
Drift	The gradual decline in an AI system’s accuracy as the world, the data or the model changes, caught by monitoring.
Rate limit and quota	Caps a provider places on how much an account may use, which features must handle gracefully.
Observability	Logging what an AI system was asked, what it answered and what happened next, so problems can be traced.
Return on automation	The honest comparison of hours and errors saved against the cost of building and running an automation, measured rather than assumed.
Pilot	A small, time-boxed deployment with a success measure, which either earns production or is stopped.

Using the glossary

What this means for you

Forty terms cover almost every AI and automation conversation a small business has. Knowing them lets you tell rules from models, drafting from deciding, and safeguards from assumptions, and to ask the questions that reveal what a pitch actually offers. The vocabulary is new; the buying discipline is not: know what it does, what happens when it fails, and who is accountable.