800 python line AI generated pong

[jgpeiro]

I created a pong game with its documentatio in about one day using Anthropic Claude 3.5 Sonnet. LLMs can have funny fails, but when works its amazing.
I started with a basic pong game, I give the API of the LCD driver and the main function. I tested multiple times and always got a working pong game, then I start to add more and more features. The game had about 20 versions, all working almost perfectly. Each iteration I test the game and provide feedback and new features for the next version. Everything worked fine, with a clear understanding of my requests and issues.

When I didnt know what more to add, I had the idea to add a pwm piezo speaker to have sounds, so I start a new conversation requesting "audio fx engine for a micropython game". While my solder was heating up, I got the first audio engine, with asyncio and able to reproduce different sounds. I soldered the piezo and iterate few times until it worked fine.

Finally I created a new conversation and past the game code and the audio engine code with the instructions like: "put all together" and it worked...So I was in a extrange situation: I had an 800 line code, more or less dense, that I cant understand. If I want to modify something manually I need to spend few time studing the code and cross fingers to modify the code in all required places, or I can simply copy to the conversation and ask to the LLM for modifications. After the see how easy was, I know that anything more or less reasonable, will work without problems...So why a company will pay me for one day of work to modify or mantain a code if an LLM can do it in less than a minute? I felt that was the end of programming...

At least I had another idea to understand the code, ask for the documentation, and I asked a markdown documentation describing all the code and iterate on it asking everything I needed to understand about the code, like how the sound engine works or how the magnet power-up where implemented. All was explained almost perfectly.

At some moment I found an error in the documentation, wich said that the code uses gc.collect() and it wasent true, so I asked for a review and it found this one and other discrepancies between code-documentation and listed all of them.

The code is uploaded here:
https://github.com/jgpeiro/ai_iterative_pong
A video can be watch in youtube:
https://www.youtube.com/watch?v=Jzg7LhgocyI
Screenshot of the documentation: