That's a rather cheap retort. I am not saying everybody should use raw untyped lambda calculus for their programming, just that we would all benefit if we could translate languages we use to and from it, because then we could interoperate with any other code, refactor it, etc.
A CIL to LC compiler is effectively an emulator of CIL in LC. That is, every primitive of CIL has a corresponding LC term, which operates on the CIL execution state. So then you can express functions from .NET standard library as terms in LC.
Now let's say you do the same for JVM. Now you can start looking at all these lambda terms and search for some similarities. For example, you might notice that some list functions are equivalent if you transform the representation in a certain invertible way. This gives you a way to express functions from one standard library in the other.
In general, I think we should try to translate wide variety of human programs into lambda calculus and then try to refactor/transform the lambda terms to spot the common patterns.
That sounds very thankless, but on the other hand we have very fast computers and maybe "the bitter lesson" of just throwing more compute at finding patterns can be applied here, as well.
