Engineer taught his dog to program and create games using AI

Former Oculus engineer Caleb Leake conducted an unusual experiment: he set up a system in which his dog, Momo, literally “programmed” games. Of course, the dog did not learn Python or C++ — she simply pressed keys with her paw, and then a generative AI took over. The results turned out to be surprisingly functional, and Leake подробно described the entire process on his blog and demonstrated it on YouTube.

How It Worked

Caleb used the chatbot Claude and configured it so that the neural network would interpret any random string of characters as “brilliant instructions from a mysterious game designer.” Momo was trained to press keys on a modified keyboard: for each press, she received a treat from an automatic feeder. The resulting meaningless sequences were sent to Claude, which interpreted them as a request to create a game.

Additionally, the AI acted as an automated tester: after generating the code, it launched the game itself and checked whether it worked. If the project turned out to be unplayable, Claude received a “redo” command and generated a new version.

All system control was handled through a single-board computer, Raspberry Pi, which connected the keyboard, the feeder, and Claude.

Minimum Requirements and Early Results

At first, the games were extremely unstable — mostly chaotic collections of objects without any logic. Leake then introduced strict criteria: the game had to include controls, sound, a playable character, and at least one enemy. After that, the system began producing fully playable projects — simple, but functional.

A video demonstrating the process and links to the resulting mini-games were published on Caleb’s blog and on YouTube.

Why He Did It

The experiment is part of a broader exploration of “vibe coding,” where a developer gives AI not strict instructions but only a general mood or random prompts, and the neural network fills in the details. Leake wanted to show how far generative models have progressed: even completely random input from a dog can be turned into a playable prototype.

It is also a humorous way to demonstrate how the barrier to entry in programming has dropped so dramatically that it is now accessible to virtually anyone — even an animal with a paw and a treat.

In Brief

Former Oculus engineer Caleb Leake built a system in which his dog Momo pressed keys with her paw, and the generative AI Claude turned random symbols into the code for simple games. After introducing minimum requirements (controls, sound, character, enemy), the system produced genuinely playable projects. Everything ran on a Raspberry Pi with an automatic feeder to motivate the dog. The experiment highlights the potential of “vibe coding” and shows how AI can now work even with completely random input. Videos and games are available on the author’s blog and on YouTube.