Why is it a problem they are at odds? Its not a researchers problem to create scalable production/consumer software or hell even git repos or libraries for you. Its yours! Thats your interest.
Doing applied research and working with people who want to do that tends to work reasonably well with production, because the point is to research techniques and algorithms and so forth that help improve production. Theoretical or more future-focused research tends to need to focus on building minimal scaffolding to be able to test ideas, when that kind of scaffolding gets integrated into production software, that’s where the problems arise, because by necessity it’s a kind of MVP. I am right now working on rearchitecting a component that was built this way and labeled as production without going through hardening first (though it did go through a great deal of testing it didn’t get the code review it needed).
In the end I suppose my point is that research and production aren’t inherently at odds, or even orthogonal. Applied research can be valuable to improving production software by studying current stable code and processes and researching improvements, but trying to do blue-sky research by incorporating code into production can be orthogonal or in some cases truly at odds with stability. It matters a great deal who is doing it, and how it’s managed.