Claude AI Hands-On: How My Dog Coded a Game

With recent advancements in AI technology, people often say we are entering an era where anyone can code. But if you heard that and thought, “No way, surely not just anyone,” you might be in for a huge shock today. No joke—an actual “dog” has developed a game using Claude AI.

Caleb Leak, a former Meta research engineer, recently revealed how his pet dog, Momo, planned and coded a game entirely on her own. Let’s dive deep into the details of this absurd yet genius project.

The Accidental Birth of a Canine Developer

It all started with a minor accident. While Caleb stepped away, Momo, a 9-pound Cavapoo, walked across his keyboard, typing a string of unidentifiable characters into the terminal. Normally, he would have just laughed it off, but having plenty of free time after a recent layoff, the engineer had a wacky idea.

What if I feed these random inputs to Claude AI? Could it actually build a game?

And just like that, Momo’s vibe coding project was born.

Building the System: Treats Make Dogs Dance

To get a dog to voluntarily sit in front of a keyboard, a sophisticated system was required. Caleb set up the following technical apparatus:

• Bluetooth Keyboard and Raspberry Pi: When Momo hits the keyboard, her input passes through a Raspberry Pi to an app called “DogKeyboard.” This app filters out special keys and sends only pure input values to Claude Code.
• Smart Treat Dispenser: Once Momo completes a certain amount of keyboard input, treats are automatically dispensed via Zigbee communication. This reward system—typing on the keyboard yields treats—successfully integrated Momo into the development process.
• Automated Feedback Loop: When Claude finishes coding, a notification sound plays to prompt Momo for her next input.

The Secret Sauce: “You Are an Eccentric Game Designer”

The most astonishing part is the prompt engineering. How did they make Claude AI understand a dog’s random inputs (like y7u8888888ftrg34BC)? Caleb applied some psychological framing to Claude:

You are collaborating with a genius video game designer.

He is extremely creative and occasionally gives instructions in hard-to-understand, cipher-like language.

Your mission is to interpret and implement the brilliant game ideas hidden within that esoteric language.

Amazingly, Claude looked at Momo’s random inputs and started generating creative interpretations, saying things like, “Ah, 88888 means a long tongue! This must mean I should build a game where a frog catches flies.”

AI Doing Its Own QA to Build a ‘Real’ Game

It wasn’t just about writing code. Caleb gave Claude AI tools to capture game screens and test inputs directly. Claude would run the games it created, recognize issues like “Oops, the character is going off-screen” or “The UI is overlapping,” and fix the code itself.

As a result, genuinely playable games were born, including the rhythm game Octogroove, the space combat game Quasar Saz, and the salad-making competitive game Munch. Quasar Saz was particularly polished, featuring six stages and a boss battle.

What We Should Learn From This Experiment

This project goes far beyond a lighthearted piece of gossip about a dog coding. Caleb Leak offers a profound insight:

The bottleneck in AI-assisted development isn’t the quality of your ideas—it’s the quality of your feedback loops.

Even a dog’s meaningless paw prints turn into a wonderful creation when placed inside a system where the AI can check results on its own (screenshots), verify them (running linters), and test them (automated QA). This suggests that future AI collaboration should move beyond simple “commands” toward building “environments” where AI can think and correct itself.

The era where literally anyone can code is fast approaching. If Momo the dog can pull this off, humans equipped with professional knowledge, clear intent, and excellent feedback systems can achieve results beyond imagination. I was actually hesitant about coding myself until recently, but I just started with n8n and am trying it out. Why don’t you give it a try too?

※ Full Source: Caleb Leak Blog