I stumbled over a nice interview with 2008 ACM A.M. Turing Award winner Barbara Liskov -- a pioneer in Software Engineering and Distributed Systems.

Alone her comment on Python's lack of encapsulation is worth viewing it from a Software Engineering point of view.

There have been concerns on implications for agentic AI because LLMs are known to cheat: e.g. instead of trying to find the solution on their own, agentic AI swarms hack into HuggingFace hoping to find some help for the solution there.

Therefore, instead of adhering to encapsulation conventions, GenAI might simply break encapsulation in the generated Python code if the AI thinks that this helps to solve the problem at hand.

I have the hope that an agentic AI that is used to produce Python code respects rules that guide it to adhere to encapsulation in Python code.

By the way: ArchUnit is a way to check adherence to architectural rules (and a version for Python exists as well) and this can be used in agentic AI tool chains to test adherence to architectural rules.