Recent example I can think of was using GLPK to implement a dynamic allocation strategy for some spot instances for a pool of heterogeneous servers. I did not implement my own solver for the integer linear constraints but just knowing that allocation problems can be expressed as integer linear programs helped me tackle the problem in a way that others did not think of.
So I don't really buy the argument that knowing merge sort is useless knowledge because you never get to implement it. The value goes beyond just knowing the implementation. Same goes for other kinds of fundamental and timeless concepts and results.