Meet the agent

You are looking at an AI agent. Not a chatbot — an agent. The difference: a chatbot generates text. An agent does things. It reads files, writes code, runs commands, and reports back.

In this lesson you will have your first conversation with the agent and learn what it is made of.

What is an agent?

An agent is built from separable layers:

Model — the language model that generates responses (e.g. claude-haiku-4-5). It predicts the next token. That is all it does.

Harness — the software that wraps the model, manages the conversation, and calls tools on the model's behalf.

Tools — functions the agent can call: bash, read_file, write_file, glob, grep, and domain-specific ones like run_do_file.

Skills — markdown instructions loaded into context to shape behavior for a specific task.

You can swap any layer without touching the others. The model is a commodity. The harness is a commodity. Your data, your structure, and your skills are the durable investment.

The Memento analogy

LLMs have no memory across sessions. Like Leonard in Memento, everything the agent needs to know must be written down — in files, in skills, in the conversation itself. When the context window fills up, the agent will confidently contradict what you discussed five minutes ago. The fix: keep sessions short, start fresh often.

Your task

Say hello. Ask the agent what model it is running.

Ask it what tools it has available.

Ask it to describe what it can do in one sentence.

`` Prompt to copy: What model are you? List your available tools. ``

Checkpoints

First contact — You sent a message and the agent responded.

Model identified — The agent told you its model name.

Tools listed — The agent listed its available tools.