AI Agent Glossary
These are the words that come up when a university, an institute or a training academy starts deciding whether to put an AI agent inside a course. Each one is defined here in plain language, without marketing, so a conversation about an agent can be had on the same terms by the people who teach, the people who buy and the people who build.
To see these terms in production, read how the Moodle AI agent answers students from a course curriculum and cites the unit and page.
AI agent
An AI agent is a software system that takes a goal or a question in plain language, decides what it needs, retrieves it, reasons about it, and carries out a task. Unlike a scripted bot, it is not choosing from a fixed list of prewritten replies.
Autonomous agent
An autonomous agent decides its own next step instead of following a fixed path. It reads what it received, chooses which action applies, and can retry, branch or stop and ask a person. Autonomous refers to how it chooses steps, not to the absence of human oversight.
Chatbot vs agent
A chatbot matches user input against a list of intents somebody wrote in advance and returns a stored reply. An AI agent reads the question, retrieves from a real data source, reasons and can perform an action. The difference shows on questions nobody anticipated, which are most questions.
RAG (retrieval-augmented generation)
RAG, or retrieval-augmented generation, is a technique where a system searches a document collection for passages relevant to a question and writes its answer from those passages. The model supplies the language, the documents supply the facts, and the answer can cite where it came from.
LLM (large language model)
A large language model, or LLM, is a model trained on large amounts of text to predict and generate language. It handles understanding and wording well. It is not a reliable store of specific facts about your business, which is why production systems give it a source to read.
Moodle
Moodle is an open-source learning management system used by universities, institutes, academies and corporate training teams to run online courses. It holds course content, activities, assessments and student accounts, and it can be extended so tools such as an AI tutor run inside the course itself.
LMS (learning management system)
A learning management system, or LMS, is the platform an institution uses to deliver and administer courses online. It stores course material, tracks student progress and manages enrolment. Moodle is the most widely used open-source LMS and is the platform E-comienza deploys course agents into.
Grounding
Grounding means restricting an AI system to answer from a defined source, such as a course curriculum or a set of financial records, rather than from whatever the model absorbed in training. A grounded answer can be traced to a specific document and checked by the person reading it.
Hallucination
A hallucination is a confident answer from an AI system that is not supported by any real source. It is dangerous because it reads exactly like a correct answer. Grounding, source citation and an explicit refusal when material is missing are the standard defences against it.
Source attribution
Source attribution, or citation, is an AI system naming where an answer came from, such as a unit and page in a course. It lets the reader verify the claim and read the full passage in context, and it lets a reviewer check the system quickly against the original material.
Multi-agent system
A multi-agent system uses several specialised AI agents, each handling one part of a task, instead of one agent doing everything. Splitting the work makes each agent easier to instruct, test and debug, and makes it clear which step produced a bad result when something goes wrong.
Prompt
A prompt is the set of instructions and context given to a language model before it answers. In a production agent the prompt is not a single clever sentence. It is engineered configuration covering the role, the rules, the retrieved material and what the agent must do when it is unsure.
Fine-tuning vs RAG
Fine-tuning adjusts a model with additional training so it adopts a style or behaviour. RAG leaves the model alone and gives it the right documents at question time. For answering from material that changes and must be cited, such as a course curriculum, RAG is usually the appropriate choice.
Vector index
A vector index stores passages of text as numeric representations so a system can find passages by meaning rather than by exact keyword. It is what lets a RAG agent retrieve the right part of a course unit when a student phrases the question in words the material never uses.
