Inside the AI Prompt Lab: How I Built a 3D Cyberpunk Game Without Coding

Hi, and welcome to The Atomic Builder! 👋

Last week was WILD! This very newsletter was hijacked by my teenage son! If you didn’t read that issue - go and check it out now, or I’ll send him round…

On to this week…

The more I build with AI, the clearer the power of communication becomes. Just as language and tone affect our interactions with each other, these things are arguably more critical when working with AI.

In a previous issue, we explored seven powerful product strategy principles I discovered while building a simple cyber parkour game for a competition I impulsively decided to enter on X - still waiting on that winners cheque...

Today, I'll show you exactly how I built that game with AI. It's not an Xbox killer - nor is it meant to be. Its creation demonstrates that these tools are opening doors for anyone willing to experiment.

I like to think of this issue as the practical "how-to" companion to the initial issue. You'll see the specific prompts I used, how I handled obstacles, and the communication techniques that bridged the gap between my creative vision and technical implementation.

Let’s dive in…

The Initial Vision

My journey began with a comprehensive prompt I extracted from Claude, to use when building my app in Cursor.

Prompt engineering is powerful, but the reality is you don’t have to ‘learn it’, per se. AI can write prompts for you. The prompt below is something you can extract from Claude (or ChatGPT etc)…You can use it as a starting point yourself.

Notice the language. There are specific terms that nudge AI in the right direction. Your role is to guide AI to do what you need. A good prompt will help you achieve that!

I’m building a 3D parkour-style game in a cyberpunk city using Three.js and JavaScript. For my initial prototype, I want:  

A basic 3D scene with a WebGL renderer, perspective camera, and simple lighting (ambient + directional).  

A small rooftop level: a flat plane (the roof) with a few cube-shaped obstacles (e.g., vents or walls) and neon-colored materials.  

A player represented by a simple mesh (e.g., a capsule or box) with basic movement:  WASD to move, Space to jump.  

Physics-based falling (using Cannon.js for gravity and collision).  

Clean, commented code in a single HTML file using Three.js and Cannon.js.

Keep it simple but extensible.

See this bit? ‘Clean, commented code’ - it’s been a game-changer for me - what it does is help you understand what’s beneath the hood of what you’re creating. If you scroll through your AI generated code it helps you build up a picture (IF YOU WANT…!) of what different functions do. Here’s what it looks like in practice:

After some back and forth, Cursor generated a working prototype. I had a functional, if pretty basic, 3D environment with a character I could control. The foundation was laid faster than it would have taken me just to set up a development environment manually (I assume).

From Box to Cyberpunk Runner

My first character was literally a rectangle... I wanted something more engaging, a little more electric:

Replace primitive shapes with a low-poly cyberpunk runner model. Ensure the existing features of the game are not impacted

The AI constructed a character styled to look like a cyberpunk runner, complete with a visor, cybernetic implants, and glowing accents. It even added animations for running, jumping, and landing - all while maintaining the physics and control mechanics. Phew!

When Things Went Wrong

Not everything was plain sailing. After adding more complex platforms, my character would get stuck on invisible surfaces. My initial attempts to describe the problem were frankly lazy, too vague and a good example of what you shouldn’t do:

The player keeps getting stuck. Fix it.

This produced changes, but the problem persisted. I knew that being (more) specific about the behavior would help more:

After the first jump, the player can't jump again. They appear to be getting stuck to the floor. When landing on a platform, the character sinks slightly into the surface instead of resting on top.

This detailed description helped the AI diagnose the issue as a collision detection problem. It implemented a more robust ground detection system that fixed the issue without breaking other game mechanics - so the character would basically know where to land.

Polishing the Experience

The game was sluggish initially. I found that adding specific visual elements dramatically improved how the game felt:

Add a simple particle trail when the player jumps that fades quickly - providing visual feedback for jumps. Ensure the trail is visible but not intrusive to the screen.

Show the total horizontal distance traveled as another metric beyond score.

I also made UI refinements with direct, specific requests. Sometimes it is worth really spelling out for AI precisely what you want:

Remove the black bar under the score. This appears to be redundant.
Move the score up to the top right next to distance and have it in the same format so there is consistency.

Even finishing touches like adding a competition badge were handled with precision:

Please add the Vibe Jam 2025 competition badge to my game. Add it in both the createStartScreen() function and ensure it persists when the game starts.
Place the badge in the bottom right corner, positioned about 40px above my existing signature.

Lessons for Non-Technical Creators

Right, where does this leave you? After completing this project, here's what you should consider in your own building projects (You can pair this with the lessons from the prior issue):

1. Start with a working prototype, then iterate. Getting something immediately was much more motivating than trying to perfect everything immediately.

2. Be specific about behaviors, not implementations. Describing what you want to experience works better than trying to specify technical solutions.

3. Use visual references and analogies. When I couldn't describe something technically, comparing it to existing games or visual effects gave the AI context. Context is king!

4. Provide context and constraints. The most effective prompts provided context and guided AI with practical constraints.

5. Debugging requires precision. When something went wrong, describing exactly what happened, when it happened, and what should have happened instead produced the best solutions.

Your AI Building Toolkit: From Game to Any Project

To help you apply these techniques to your own projects, I've created a practical resource you can use right away. It’s available exclusively to you at the Atomic Builder Hub:

🚀 VIBE CODING STARTER KIT: The AI-Powered Builder's Playbook

This one-page cheat sheet distills lessons I've learned about effectively collaborating with AI into a ready-to-use template with examples to get you thinking in the right way.

GET THE VIBE CODING STARTER KIT

Final Thoughts

When creating products using AI tools, what matters now is having a clear vision and the ability to communicate it effectively - skills that creators already possess.

For non-technical creators, this means the question is no longer "Can I build this?" but rather "What exactly do I want to build and how do I guide the AI to do this?"

The future of digital creation belongs to those who can imagine and communicate their vision clearly. And that's something anyone can learn.

I'd love to see what you create. Share your projects or questions - just hit reply.

Now, I'm off to finish CYBERPUNK PARKOUR. In stores Christmas 2025!!! (Not really...)

Until next time, keep experimenting, keep building, and as always - stay atomic. 👊

Faisal

This Week’s Build Beats 🎵

Each issue, we pair the newsletter with a track to keep you inspired while you build.

This week, because AI building bridges the gap between imagination and implementation…!

🎧 "Blinding Lights" – The Weeknd

Grab the playlist on Spotify - I add to it each week!