AI Agents: Models That Use Tools
The big idea: a plain chatbot can only talk. It reads your words and writes words back. It cannot look anything up, run a calculation it is unsure about, or actually do something in the world. An AI agent changes that. An agent is still a large language model (LLM) at its heart, but we give it three extra things: a set of tools it can call, a goal to achieve, and a loop that lets it keep working step by step until the goal is done.
In short: Agent = LLM + Tools + a Goal + a Loop. This one lesson is your complete textbook on how that works, why it works, where it is used in real products, and how it goes wrong. Read it slowly; every new word is explained the first time it appears.
What you will be able to do after this lesson
- Explain the difference between a model that answers and an agent that acts.
- Describe function calling (also called tool use): how a model asks for a tool and how the system runs it.
- Walk through the agent loop — plan → act → observe → adjust — and trace a real multi-step task through it.
- Recognise real agent products: coding assistants, deep-research modes, and computer-use agents.
- Name the common failure modes (loops, wrong tool calls, compounding errors) and the safety practices that keep agents in check.