I haven't read the article (because I hate Medium) but I reckon the biggest reason why LLM-assisted projects use Python is because there is a metric buttload of python code on the web to be slurped up and used as training fodder.

Now I'm curious: is a metric buttload much larger than an imperial one?