but using Python throughout, or even Python with a sprinkling of Cython/C++/C for performance, allows for cleaner and faster engineering than using a special language for each niche.
I don't think that R has a bigger problem with non-programmers being bad software engineers than Python has - there are plenty of people who know Python passably and are quite happy that they can be productive without being good software engineers (versus Java where the intent of the language is biased for everyone to write code with a minimum quality standard rather than everyone to write code with focus on being productive). But you can find decent Python software engineers, and more recently you can find decent software engineers who also know enough of the niche in question to produce high-quality production code in that niche from the get-go rather than throwing models over a data science - engineering wall that exists between two departments.