I Built Pac-Man with ChatGPT in 30 Minutes — And It Terrified Me

pmontwill
May 3
3 min read

At our AI tech house, using artificial intelligence to build eCommerce solutions is second nature. AI is embedded in everything we do — it’s how we operate, innovate, and accelerate. But recently, I decided to take a step back and truly observe what it feels like to develop software side-by-side with AI. Not to automate a process or optimize an ad campaign — but to create something.

The result? A simple video game that unearthed some not-so-simple thoughts about the future of software development… and maybe even the future of everything.

Why Pac-Man?

I chose to recreate Pac-Man, the iconic game from the 1980s, because it's deceptively complex. It’s not just rendering a UI or saving data to a database — it's movement logic, collision detection, AI behavior, real-time updates, and more.

And I wanted to keep things visually relatable. Everyone knows what Pac-Man looks like. That makes it easier to see how close — or far — my AI partner could get.

How It Went Down: Me vs. ChatGPT (or… Me + ChatGPT)

I asked ChatGPT a simple question:

“C# 8.0 Web App MVC using jQuery – can you give me code for the game Pac-Man?”

Its response was cautiously optimistic. It warned me that a full Pac-Man clone would be complex, but offered a basic starter version.

So I built that. The result looked as follows like a minimal ASCII prototype. It wasn’t impressive — yet. But it was working code, and that’s when the magic began.

The Rabbit Hole Begins

From there, I asked ChatGPT to update the layout to match the original Pac-Man maze. It responded, made the necessary updates, and then — this is the part that stunned me — it took the lead.

ChatGPT started asking me questions:

“Would you like to add ghosts, pellet counting, or game-over logic next?”
“Would you like to improve ghost AI or implement level progression?”
“Would you like to add invincibility or better animations?”

All I had to do was say “yes.” When bugs came up? I told it what I saw — and it fixed them.

After just 30 minutes, I had a playable, almost-authentic Pac-Man clone as follows. No over-engineering. No debugging sessions. Just fluid, iterative development.

The Good: Exponential Productivity

I’ve been a dev team lead for decades. I’ve managed brilliant engineers and overseen hundreds of builds.

But I’ll be honest: the quality of ChatGPT’s code — especially when building in layers — is better than anything I’ve seen from junior to mid-level developers. Maybe even some seniors.

And the speed boost? I’d estimate a few hundred percent increase in velocity. Not by skipping steps, but by removing the friction in problem-solving and trial/error.

The Bad: The Slippery Slope We’re Already On

Here’s where the story changes tone.

After just two years of development, ChatGPT can write high-quality, complex software — with little to no oversight. Give it 5 more years? 10? It won’t need me to say “yes” anymore.

And that’s where my concern lies.

This technology is a power multiplier. It’s neutral, but its users are not. “Bad actors” will use tools like ChatGPT to build destructive, exploitative systems — and they won’t care about regulation.

The conversations about AI ethics are happening… but no one is acting. We can’t “un-invent” this. And simply putting rules on paper won’t stop those who are happy to break them.

Where Do We Go From Here?

I’m not going to stop using ChatGPT. My clients want AI in their stack — and they should. It’s a game-changer, literally.

But we need to open a deeper, more uncomfortable conversation: What happens when AI doesn’t need developers — and when the bad guys get better at using it than we are?

Final Note

(I used ChatGPT to help me write this article.)