A guided learning experience · Paul Cheek · MIT

Build your
first AI agent.

No experience required. A guided, interactive skill — created by Paul Cheek, MIT Senior Lecturer on AI in Executive Education — teaches you what an agent is actually made of, asks you questions to shape your own, and hands you working code at the end. Then you run it.

Isometric illustration of an AI agent architecture
01Learn the parts. The skill teaches every component of an agent, in plain language.
02Answer a few questions. You design your agent through a guided conversation.
03Get the code. Walk away with a working agent you can run yourself.
How the skill works

A teacher, not a template.

This is not a fill-in-the-blanks generator. The skill installs into Claude and turns the conversation into a lesson. It introduces one concept at a time, checks your understanding, and uses your answers to assemble an agent that is genuinely yours.

// it teaches

It explains the
concepts

Before you build anything, the skill walks you through what an AI agent is and the pieces that make one work — so you understand what you are assembling, not just copy it.

// it asks

It interviews
you

You are asked focused questions — what your agent is for, what it should be able to do, how it should behave. Your answers become the specification for your first agent.

// it delivers

It hands you
the code

At the end, the skill generates a working agent from your decisions and shows you exactly how to run it. You leave with something real, and the understanding to change it.

What you'll learn

Inside every agent.

By the time you finish, these five ideas will feel obvious. The skill introduces them one by one, then shows how they fit together into a single working system.

01

The Goal

What the agent is for — the job it exists to do and how you'll know it worked.

02

Instructions

The agent's standing orders: how it should think, behave, and respond.

03

Tools

What the agent can actually do — the actions and data it can reach beyond text.

04

Memory

What the agent knows and remembers — the context it carries through a task.

05

The Loop

How it ties together: reason, act, observe the result, and decide what's next.

Get started · three steps

Three steps to your
first agent.

Work top to bottom. Each step lights up when the one before it is done.

1

Download the skill

Grab the skill file and save it somewhere you can find it — your Downloads folder is perfect. It's a single .skill file.

Download the skill → — Downloaded
2

Install it in Claude

Open the .skill file you just downloaded. How you install it depends on where you use Claude.

Claude desktop app Easiest

Just double-click the downloaded .skill file. The Claude desktop app opens and installs it automatically. That's it.

Claude on the web claude.ai
  1. Go to claude.ai and open Settings → Capabilities (Skills).
  2. Choose Upload skill and select the .skill file you downloaded.
  3. Toggle the skill on, then start a new chat so Claude can use it.
— Installed
3

Paste the kickstart prompt

Open a new chat in Claude, copy the prompt below, and paste it in. The skill takes over from there and starts teaching.

I just installed Paul Cheek's "Build Your First Agent" skill, and this is my first time building an AI agent. Be my guide.

Teach me as we go: explain each component of an agent — the goal, its instructions, its tools, its memory, and the loop that ties them together — and ask me one question at a time so the agent we design is genuinely mine.

When we're done, give me code I can run, and tell me exactly how to run it. Let's start.
— Copied · paste it into Claude
Before you go

What did you just learn?

Building the agent is half of it. Naming what changed in your understanding is what makes it stick. Take a minute.

// unlocks once you've copied the kickstart prompt above
What do you want to learn next?
Saved. You've built your first agent and named what you learned — that's exactly how this is supposed to feel. Bring your "next" idea back into Claude and run the skill again.