If all the libraries are rust as the article claims having the top layer in Python probably makes even less difference.

I tend to agree with the article’s statement about the value of the test code though, may even have been true before LLM code took over.