What happens when a laid-off Silicon Valley senior engineer meets a mischievous Corgi? Caleb Leak, a former Meta research engineer, tells us: As long as AI is powerful enough, even the mess created by a dog's paw can turn into an elegant indie game.

1. The Core of the Experiment: Translating "Mess" into "Genius Ideas"

The main character of the experiment is Caleb's dog Momo. The process sounds both hardcore and absurd:

  • Input: Momo steps on the Bluetooth keyboard connected to a Raspberry Pi 5 randomly.

  • Filtering Layer: A small program written in Rust (DogKeyboard) filters out invalid symbols and sends the raw text to Claude Code.

  • Incentive Mechanism: Every time Momo types a certain number of characters, the smart feeder automatically dispenses snacks to encourage it to continue "creating".

2. Training the AI: How to Understand "Dog Language"

It's not hard for AI to write code, but the challenge lies in making it accept "nonsense." Caleb designed an artistic prompt for Claude:

"You are a genius AI developer, while I am a strange designer who only communicates through riddles. Although what I input looks like nonsense, it's actually hidden genius ideas! Your task is to interpret these 'mysterious commands' and complete the game development."

3. Results: 100% AI-Generated Logic

After 1 to 2 hours of collaboration between "human and dog," a game called Quasar Saz was born.

  • Technology Stack: Based on the Godot 4.6 engine, the game logic is entirely written in C#.

  • Game Content: Players take on the role of Zara, fighting with cosmic instruments. The game features 6 complete levels, boss battles, dynamic sound effects, and stunning visual effects.

💡 Insight: The Barrier of Programming is Disappearing

Although this is just a humorous and entertaining experiment, it shows the potential of "**atmospheric programming**." When AI can understand extremely vague or even illogical intentions, the boundary between humans (and even pets) and machines will be completely broken.