Part 7 of 9 · 1 min read
Words you will meet
A short glossary of the terms this path uses the same way on every page.
These words come back again and again. They always mean the following.
| Word | Meaning in this path | Example |
|---|---|---|
| Language model | A program that predicts the next text | Continuing "The capital of France is" with "Paris" |
| Tool | A function the model can ask your program to run | look_up_order(4471) |
| History | The record of everything asked and done so far in a run | Question, tool call, result, tool call, result |
| Loop | Ask the model, run its tool, add the result, ask again | Repeating until the refund is decided |
| Limit | A rule that stops the loop | At most 8 steps |
| Workflow | Your code fixes the steps | Look up, check, reply |
| Agent | The model chooses the steps | "Sort out this refund" |
| Agentic AI | Building software where models act toward goals with limits and checks | A refund assistant with a human check |
| Autonomy | How much the system decides without a person | Approve every step, or run alone |
| Multi-agent system | Several agents that hand work to each other | Researcher, writer, checker |
| Prompt | The text you give the model | "You are a polite support assistant." |
| RAG | Retrieval-augmented generation: look up relevant text, then give it to the model | Find the refund policy, then answer from it |
| MCP | Model Context Protocol: a shared way for tools to talk to agents | One standard plug for many tools |
Where each word is taught
Prompts are in Level 3. RAG is in Level 4. The loop and limits are in Level 5. Tools and MCP are in Level 6. Multi-agent systems are in Level 8.