AI glossary for people outside the technical world
The terms you will come across when talking about artificial intelligence and automation, explained in clear language, with no technical templates.
AI agent
An AI agent is a system that does not just answer questions, but can make decisions and carry out actions on its own to meet a goal: querying a database, sending an email, updating a CRM or chaining several steps without anyone telling it each one individually. The difference from a simple chatbot is that ability to act on your systems, not just to converse. When we talk about custom AI agents at AIGiner, we mean this kind of assistant built specifically for a concrete process in your company.
AI Act (European Artificial Intelligence Regulation)
The AI Act is the European Union regulation that governs the use of artificial intelligence according to the risk each system poses: the higher the risk to people, the more obligations around transparency, documentation and human oversight it requires. It applies to any company that develops or uses AI within the EU, including SMEs that integrate chatbots, automations or AI-powered analysis tools. Its obligations apply progressively, so it is worth checking what timeline applies to each requirement depending on the type of system you use.
Hallucination
Hallucination is the term for when an AI model generates information that sounds convincing but is false or made up, without the model itself "knowing" it got it wrong. It happens because these models generate the most probable response based on what they learned, not because they check a real-time source of truth. It is one of the reasons why, for important processes, it is worth combining AI with human review or with techniques like RAG, which force it to rely on real documents.
API
An API (application programming interface) is the mechanism that lets two programs communicate with each other in an orderly way: one asks for something and the other responds, following rules known to both. Almost every tool your company uses, from the CRM to accounting software or WhatsApp, exposes an API, and that is precisely where most of our automations connect. If a tool has an API, it can almost always be integrated with the rest of your operations.
Automation
Automation means a system carries out repetitive tasks without human intervention whenever certain conditions are met: an email arrives, a form gets filled in, a value changes in a spreadsheet. AI does not have to be involved, although the two are increasingly combined: automation moves the data and executes the actions, and AI provides the judgment to decide what to do with the cases that do not follow a fixed rule.
Chatbot
A chatbot is a program that holds a conversation with a person, usually by text, to answer questions or resolve simple requests. Chatbots based on generative AI understand natural language and hold a fluid conversation, unlike the older decision-tree chatbots, which only recognized exact phrases or preset buttons. They can live on your website, WhatsApp or social media, and escalate the conversation to a person when the case requires it.
Embedding
An embedding is the numerical representation of the meaning of a piece of text, an image or any other content, so a computer can compare it mathematically with others. Two sentences with a similar meaning have similar embeddings, even if they use different words. It is the technical foundation that lets a system search by meaning instead of by exact word matches, which is essential for AI search and for RAG.
Fine-tuning
Fine-tuning is the process of adjusting an already-trained AI model with specific examples from your business, so it responds in your tone, your terminology or your way of resolving particular cases. Unlike writing a good prompt, which only changes the instruction for a single query, fine-tuning modifies the model itself in a more permanent way. It usually comes up when a well-designed prompt is no longer enough to achieve the consistency you need.
LLM (large language model)
An LLM (Large Language Model) is the kind of AI model behind tools like ChatGPT or Claude: a system trained on huge amounts of text that learns to predict and generate language coherently. It does not "know" things in the human sense, but it recognizes patterns in language with remarkable precision. It is the engine that lets a chatbot understand a question phrased in any way and respond naturally.
MCP (Model Context Protocol)
MCP is an open standard that lets an AI model connect securely and in an orderly way with external tools, databases or applications, without having to build a separate integration for every case. It works as a common language between the model and the rest of your software: instead of building a custom bridge for each tool, the model "speaks" MCP and accesses what it needs. It is a key building block for creating AI agents that genuinely act on your systems, not just converse.
NLP (natural language processing)
NLP is the branch of artificial intelligence concerned with a computer understanding, interpreting and generating human language: analyzing the sentiment of a review, extracting data from a contract or classifying an email by its content. LLMs are the most powerful evolution of NLP, but the term also covers simpler, more specific techniques that have been in use for decades.
On-premise
On-premise means a system runs on the infrastructure owned by the company itself, on its own servers, instead of on an external provider cloud. It is the option that gives the most control over where data lives, and it is often relevant for companies with strict confidentiality or regulatory requirements. Many automation platforms, for example, can be deployed on-premise because the software is the same whether the client hosts it or a provider manages it.
Prompt
A prompt is the instruction or question you give an AI model to get a specific response. How you phrase that prompt, what context you include and what examples you provide greatly influence the quality of the response: asking a model to "write an email" is not the same as giving it the exact context, tone and goal you need. Designing good prompts is, in practice, a core part of building any AI solution.
RAG (Retrieval-Augmented Generation)
RAG (Retrieval-Augmented Generation) is a technique that makes an AI model search your own documents first before generating a response, instead of answering only with what it learned during training. That way, the response is grounded in real, verifiable information from your company, and it can usually cite the specific source it came from. It is the foundation of what we call AI search at AIGiner: asking a question in natural language about your own documents and getting an answer with the source cited.
GDPR applied to AI
GDPR is the European law that protects personal data, and applied to AI it requires you to know what data feeds a model, where it is processed, under what legal basis, and that people can exercise their rights (access, rectification, deletion) even when the processing is carried out by an automated system. It is especially relevant when using chatbots, data analysis or automations that handle customer or employee information, and it is worth reviewing before putting any solution into production.
Token
A token is the smallest unit of text an AI model processes: it can be a whole word, part of a word or even a punctuation mark. AI models measure their capacity in tokens, both the ones they receive and the ones they generate, and each model has a maximum limit it can handle at once, which determines how much text or context you can give it in a single query.
Vector
In the context of AI, a vector is a list of numbers that represents the meaning of a piece of content (text, image, audio) so it can be compared mathematically with others. It is the result of calculating an embedding, and it is stored in vector databases specialized in finding, among thousands or millions of vectors, the ones most similar to a given one. It is the technology that makes it possible to search by meaning instead of by exact word.
Webhook
A webhook is an automatic notification that one application sends to another the exact moment something happens: an order is received, a form is filled in, a payment status changes. Instead of a system constantly having to check whether something happened, the webhook notifies it instantly, which makes it the most common trigger for starting an automation in tools like Make or Zapier.
Workflow
A workflow is the sequence of steps an automation follows, from the event that triggers it to the last action it executes. On visual automation platforms, a workflow is represented as a chain of connected nodes: each node does one thing (read a piece of data, call an AI, update a system) and the complete flow defines the automated process from start to finish.