Understand · Glossary

The AI glossary, in plain English.

Every term you’ll hear, defined in a sentence, with an example from a small business. Terms are linked wherever they appear.

From the glossary

Agent

An AI that takes actions toward a goal, not just one that answers questions.

An AI system that can plan steps and use tools (like your email, calendar or a browser) to get something done, checking in with a person when it should.

Why it matters

Agents can take whole chores off your plate, but because they act, their mistakes land in the real world. Decide what they may do alone and what needs your approval.

At work

An inbox agent sorts new inquiries, drafts replies and suggests meeting times, then waits for your approval before anything is sent.

Think of it like a capable new assistant: give it a goal and access, and it gets on with it, but you still sign off on the important things.

Where it appears

Suggest an edit

BiasWhen an AI’s output is skewed or unfair, usually because of patterns in the data it learned from.

Fine-tuningExtra training that adapts an existing model to a particular task, format or tone.

GuardrailsRules and limits that keep an AI tool within safe, agreed boundaries.

InferenceA trained AI model at work: taking your input and producing an answer.

Knowledge cutoffThe date a model’s training data ends. It knows little or nothing about what happened after.

LLMLarge language model: an AI trained on huge amounts of text to read and write language.

Neural networkLayers of simple connected units that learn patterns from examples: the basic design of modern AI.

Open weightsA model whose files you can download and run on your own machine.

RAGLetting an AI look things up in your own documents first, so it answers from them, not from memory.

System promptBehind-the-scenes instructions that set how an AI tool behaves before you type anything.

Voice AIAI that listens and talks back: it understands speech, replies aloud and can handle a phone call.