AI Glossary: 30 Terms Explained in Plain Language

Key Takeaways

• Most AI terms come down to a few ideas: a language model that produces text, and the ways of connecting it to your data and systems.
• For company knowledge the key terms are RAG, embeddings and vector databases; for processes they are agent, workflow and automation.
• Hallucination, prompt injection and guardrails describe the risks and how to contain them.
• In Switzerland the existing data protection act applies to AI, and the EU AI Act can affect Swiss firms too.
• You don't need to know every term. It matters more to keep asking providers until you understand what a system does with your data.

This glossary explains the 30 AI terms you're most likely to come across when talking to providers, reading about AI or planning a first project. Each one gets a few sentences in plain language, along with what it means for a small business. The order runs from the basics through working with AI to agents, risks and rules.

The basics: what's behind AI?

1. Artificial intelligence (AI)An umbrella term for systems that do tasks that used to need human thinking: understanding language, recognising images, finding patterns, writing text. When people in an office talk about AI today, they usually mean language models.
2. Machine learningThe part of AI in which a system learns from examples instead of having every rule programmed. Spam filters, fraud detection and forecasts have worked this way for years.
3. Generative AIAI that creates new content: text, images, audio or code. ChatGPT, Claude and Gemini are generative AI for text, and image generators create pictures from a description.
4. Large language model (LLM)A model trained on vast amounts of text to understand and produce language. It's the core of most AI assistants. An LLM writes the most likely continuation of a text and doesn't look anything up while doing so.
5. TrainingThe process in which a model learns from data. Large language models are trained by their providers. When you use a tool, you normally aren't training the model. What matters to you is whether the provider may use your inputs for future training.
6. TokenThe unit in which a language model processes text, usually a word or part of a word. Model use is often billed per token, and limits such as the context window are measured in tokens too.
7. Context windowHow much text a model can take into account at once: your question, any documents you've added, the conversation so far and the answer. Whatever doesn't fit, the model doesn't see. Today's models often hold several hundred pages, but not your whole company archive.
8. Knowledge cut-offThe point in time up to which a model's training data goes. The model only knows about anything later if it searches the web or you give it the information. It never knew your internal data in the first place.

Working with AI: which terms come up every day?

1. PromptThe instruction or question you give an AI. Good prompts state the goal, the context, the format you want and any limits. A system prompt is a standing instruction in the background that applies to every conversation, such as an assistant's tone and rules. More in prompting basics.
2. HallucinationAn answer that sounds convincing and is wrong, such as an invented source or a wrong date. It can't be prevented entirely, but sources and checks keep it well contained. See why AI makes things up.
3. TemperatureA setting that controls how much randomness goes into a model's wording. Low values give more consistent, predictable answers, high values more variety. Low values suit facts and data, higher ones suit brainstorming.
4. AI assistant or copilotAn AI tool you work with in conversation, such as ChatGPT or Microsoft 365 Copilot. It helps with writing, summarising and searching, while the decision and the next step stay with you.
5. MultimodalA model that handles images, audio or PDFs as well as text. In practice that means it can read a photo of damage, a sketch or a scanned form.
6. EmbeddingA series of numbers that represents the meaning of a passage of text. Texts with similar meanings have similar embeddings. That's how a search finds relevant passages even when they use different words.
7. Vector databaseA database that stores embeddings and finds the most similar ones very quickly. It acts as the memory of many search tools for company knowledge.
8. RAG (retrieval-augmented generation)The AI first looks up the relevant passages in your documents and then writes its answer from them, with sources. It's the usual technique behind AI search across company knowledge, explained in detail in RAG explained.
9. Fine-tuningTraining an existing model further on your own examples to sharpen its style, format or a narrowly defined task. For company knowledge that changes, RAG is usually the better route.
10. Open-weight modelA model whose trained weights have been published, so you can run it on your own infrastructure. It's often called an open-source model, even when not everything about it is open. It gives you control over where data is processed, but needs know-how and computing power.
11. APIAn application programming interface, through which two systems exchange data. APIs are how language models are built into your software and how an automation reaches your CRM, accounting or file storage.

Agents and automation: what do the terms mean?

1. AI agentA system that uses a language model to work towards a goal in several steps, uses tools to do so and chooses its own route within the limits you set. More in what is an AI agent?
2. Tool use (function calling)A model's ability to call a tool instead of only writing text: run a search, check a calendar, create a record. Every agent is built on it.
3. MCP (Model Context Protocol)An open standard that Anthropic introduced at the end of 2024 and that many major providers now support. It defines how AI applications connect to tools and data sources, a bit like a universal plug.
4. WorkflowA process made of set steps, for example from an incoming inquiry to the quote that goes out. In AI development, a workflow often means a process whose steps are fixed in advance, as opposed to an agent that chooses them itself.
5. Rule-based automationAutomation on the principle of “if this, then that”, for example with workflow tools such as Zapier, Make or n8n. Reliable and cheap as long as the inputs have a fixed form. When AI should come in is covered in AI or rule-based automation.
6. RPA (robotic process automation)Software that imitates clicks and keystrokes in a user interface. Useful for older systems without an interface for other software, but fragile whenever the screens change.
7. Human in the loopA person is built into the process and checks or approves before things move on. For almost any AI solution in a small business, it's the right way to start.

Risks and rules: what should you know?

1. Prompt injectionAn attack in which hidden instructions in a text, an email or a web page try to get the AI to do something other than what you want. It's a particular concern with agents that read email and are also allowed to act. Narrow permissions and approval by a person are the protection.
2. GuardrailsRules and technical limits that set what an AI may do: which data it sees, which topics it avoids, when it hands over to a person. For a team, clear usage rules are part of it too.
3. FADP (Swiss data protection act)The Swiss Federal Act on Data Protection, in force in its revised form since 1 September 2023. According to the Federal Data Protection and Information Commissioner (FDPIC), it's worded in a technology-neutral way and therefore applies directly to AI. More in AI and the Swiss FADP.
4. EU AI ActThe EU's regulation on artificial intelligence, in force since 1 August 2024, with obligations that apply in stages. It also covers providers and users outside the EU when they place AI systems on the EU market or use them in the EU. What this means for Swiss firms is set out in EU AI Act for Swiss companies (as of October 2026).

You don't need to know all of these terms to get started with AI. It helps more to keep asking each provider until you understand what a system does with your data. If you'd like to find out where AI would help your business most, our page on where to start with AI is a good next step. The entries on the FADP and the EU AI Act are for orientation and are not legal advice.