[ActorNightly]

It honestly depends on what you work on for your argument.

In general, think about the general use case for all types of development. For example, I have a bunch of finance code. I host it on a Jupyter Lab server in notebooks, so I can run/edit it from anywhere.

If I want something that is fast (which in this case is rare to see outside of using pytorch), I can write a C extension quite easily - the amount of effort that it takes me to set up boilerplate code is the same amount of effort as it takes thinking about Rust memory semantics. The code will be quite isolated from the rest of the system (unless you purposefully try to break Python), and you don't really have to even do dynamic memory if you are trying to do processing. I did this at work with fast json parsing and data extraction, and it was super easy to do.

Then there is the whole ML aspect, which is first and foremost python before anything else. Especially in the whole data sets, with the fact that you can easily generate training data in Python, whereas other languages require a lot of custom code.