Me too. Training a neural net of this scale is basically a lossy compression and promting just an interpolation of the data. But explain this to lawyers and courts...
There are so many options to emded R in any kind of system. Thanks to the C API, there are connectors for any if the traditional language. There is also RServe and plumber for inter-process interaction. Managing dependencies is also super easy.
My employer is using R to crunch numbers enbeded in a large system based on microservices.
The only thing to keep in mind is that most people writing R are not programmers by trade so it is good to have one person on the project who can refactor their code from time to time.
Unlike many other languages, R has a native/built-in tabular data structure. So when your data have tabular structure R is by far the best glue for building pipes between external libreries. If the data fits in RAM it literally doesn't have to leave the data.table object throughout the whole pipeline (including all the cleaning and transformations).
The only meaningful alternative I see is Python with maybe Polars or DuckDB.
We use embedded R in production in a way some other companies would use Python and I can say having a better compiler would definitely help.
Even if most people use R interactively, having contributers working on compiler has many positive spillovers for the language.
Also note that the R code running behind the scenes of your scripts (powering the functions of your favourite packages) is quite a different language, using less dynamic features. This is where a better compiler would always be appreciated.