Welcome — and your first build
The definitive AI coding course opens the way it means to continue: not with a syllabus, but by building something real in the next ten minutes. Buckle up.
01Instant gratification, on purpose
Most courses begin with objectives, introductions, and logistics. This one doesn’t. The very first thing you do is roll up your sleeves and build something. There will be plenty of time for theory later — and Class 3 is full of it — but the fastest way to understand what AI coding agents can do is to point one at a blank folder and watch it work.
Every project in this course is chosen to have real commercial value. Today is the single exception. Today we just have fun — because there’s no better way to feel the power of these tools than to build a game out of nothing.
02Meet Cursor
We start with Cursor — hugely popular, easy to use, and it has a free trial. If you’ve already burned your trial, you can simply watch, or follow along in any other agentic tool. Cursor is built on top of VS Code, so if you’ve used that editor, the windows and menus will feel familiar.
Download & install
Go to cursor.com and grab the build for your platform — Mac, Windows, or Linux. On Windows, accept the agreement and keep the default that adds Cursor to your PATH. On Mac, drag it into Applications.
Sign in or sign up
First launch prompts you to sign in. Create a free account if you don’t have one — the trial is all we need for now. Can’t see the screen? Use File → New Window.
Open a project
“Open a project” really means choose a folder. Make a projects folder,
create a new folder inside it called Instant, and open it. When you see
INSTANT in the top-left, you’re in.
03Build a game from one sentence
On the right-hand pane you talk to an LLM — your coding sidekick. Below the box you can pick a model (on higher plans; otherwise it’s set to auto and Cursor chooses). Then you type your request:
Please build a website for a 3D first-person shooter game in an arena against one computer opponent, controlled by arrow keys and space bar to shoot.
Send it. The agent springs to life: planning, exploring, then writing files. You watch
index.html appear, maybe a stylesheet, maybe a script. It might ask permission for
certain actions — say yes if you’re happy — or it might just go, depending on your settings.
When it finishes, open index.html in your browser. Arrow keys to move and turn,
space to shoot. A playable first-person shooter — built from a single sentence, with zero code written
by you.
Now make it better — by talking
The real lesson is iteration. Keep the conversation going and the agent revises the same files:
“That’s great. Please add some detail to the opponent so it looks more like an enemy.”
“Please add a heads-up display (HUD) and also make the difficulty harder.”
A tougher game with a HUD and a meaner enemy — no code, just plain-English feedback.
Run this ten times and you get ten different games — sometimes one file, sometimes three; sometimes the shooting
works first try, sometimes it doesn’t. That unpredictability is real, and it’s the whole point of Class 3’s exercise.
If something breaks, just tell the agent what went wrong, or delete the Instant folder and
start again with a different model or prompt.
04A teaser of going pro
To show where this is heading, the same prompt was run through a technique called a Ralph Loop with Claude Code — set off zero-shot (one instruction, no human feedback) and left to run. The result was a polished shooter with a health system, a live mini-map, kill counter, and a genuinely cool weapon. You’ll be able to do this yourself within days. We’ll cover Ralph Loops properly on Class 3.
✓ Key takeaways
- The fastest way to understand agents is to use one — build first, theorize later.
- An agent takes a goal in plain English and plans, writes files, and iterates for you.
- Non-determinism is normal: the same prompt yields different results every time.
- You steer entirely through conversation — describe what you want and what’s wrong.