I did a "Vibe Code Exploration" recently. Code was easy to generate and have kind of work. Images / assets where a challenge, but not too bad. Audio was the hardest. Claude for code, chatgpt for image gen, audiocraft & elevel labs for audio: gabormakesgames.com/Prototypes/B...

Some other ai gen exploration, i used claude to finish a half abandoned text editor i was working on: gabormakesgames.com/Prototypes/C... Claude was capable enough to build the MDI including tab docking. Pretty decent.

This is another build that i added a tutorial and Poki SDK integration to (well AI did not me).
poki.com/en/preview/8...
Prompt to add tutorial: gist.github.com/gszauer/12fe... Working on localization, but it exceeded my token limit :p

Has anyone made the gambling comparison? Feels like i'm gambling. Write a prompt, wash some dishes while it runs, check on it, rinse and repeat. I had a "just one more prompt" feeling before putting it down last night, wild.

Mine is C++, i spent a bit of time having to get things to build, and assembling libraries. I had the AI make a C++ version, but it kept getting confused. Unity version was a walk in the park tough. Tried the same thing with Deepseek, ChatGPT and Gemini, gemini was the most correct most of the time.

IDK, would a blog or something be interesting?

I've tried the same prompts with a few different models, and by far DeepSeek has been the best at writing code that mostly works. Chat GPT with 03-mini-high is runner up. Both thinking hardest :p