# Understand AI agents by building one.

Open the kit in **Claude Code** and ask *“What do I do next?”* Claude Code writes a small AI agent in front of you and explains each step. You type no code.

[Download the Agent Learning kit](https://kit.peaceandai.org/agentlearn/downloads/agentdev-kit-0.6.1.zip)

v0.6.1 · ZIP · 100 KB · 2026-11-09

## Three steps, then Claude Code takes over.

1. **Download and unzip the kit**

   Mac: double-click the zip. Windows: right-click it, then **Extract All**.

   Keep the **agentdev-kit** folder somewhere easy, like Documents. Your agent will live there. (The folder keeps its old name for now, so copies people already have keep updating.)
2. **Open the folder in Claude Code**

   In the Claude desktop app: the **Code** tab → **Local** → **Select folder** → agentdev-kit. You need a [Claude subscription that includes Claude Code](https://code.claude.com/docs/en/quickstart) (Pro, Max, Team or Enterprise).

   If asked whether you trust the folder, say yes.
3. **Ask what to do next**

   Type:

       What do I do next?

   It checks your computer, helps you get a key, and starts building.

## Understand how an agent works, and make better decisions about any agent you build or buy.

You start with one model call and end with an agent, built by Claude Code in front of you. On the way you see what a call costs and how open a model really is; who instructs the model, and why that matters; what makes an agent trustworthy, and what protects it; and who controls each part of your setup: the model, the harness and the runtime. It applies to every agent you work with.

The agent runs on a practice website we made up, so nothing real is at risk. [See the made-up practice site.](https://kit.peaceandai.org/agentlearn/lab/)

**Read it before you download.** Everything your agent is told, as plain text: [the whole kit on one page](https://kit.peaceandai.org/agentlearn/kit.md), or one part at a time: [the instructions](https://kit.peaceandai.org/agentlearn/kit/CLAUDE.md), [agentdev-journey](https://kit.peaceandai.org/agentlearn/kit/skills/agentdev-journey/SKILL.md). Give this page’s address to Claude or ChatGPT and it can read them too.

## Two accounts.

- **Claude Code**: Your Claude subscription. It writes the code.
- **A near.ai account**: Pay as you go. It runs the agent you build. In our two test runs (2026-10-10) the agent’s model calls cost about a cent or less per run, from the costs shown in the step footers, and Claude Code worked for about 40 minutes through all of Part 1. That is a measurement, not a promise for your session.

Claude Code walks you through getting the key. It keeps your place in **PROGRESS.md**: stop any time, pick up later.

## Coming to the workshop?

Get to **“Ready for the workshop”** first. Open the kit and say hi.

[Download the Agent Learning kit](https://kit.peaceandai.org/agentlearn/downloads/agentdev-kit-0.6.1.zip) · [See the workshop slides](https://kit.peaceandai.org/agentlearn/slides/)