Also, do not be deceived by "data science": it's mostly a bullshit term, and translates roughly to "programmer who knows basic statistics"
This is utter rubbish; I really wish people would keep quiet about things they know nothing about. I suggest that you have never actually discussed a domain with a data scientist if you think it's "basic statistics".
In our dev. shop, we have a lot of great programmers, but none of them can touch our data scientist when it comes to working out what our tens of millions of users are actually doing and what their salient attributes are.
As for the data scientist needing to 'implement their theories', that's what the developers are for. The data scientist does the analysis, then works with the developers to implement systems that incorporate the results. Neither group is capable of the other's work.