Seems like system design obviously requires great expertise in the field -- the designers must clearly understand what is tedious to check, what needs to be correct for general programs of a certain archetype to work, and provide the minimum amount of infrastructure to allow for robust and fast computations of the required work, while not overengineering or sacrificing generalizability.

From what I see, Domain-specific languages can be a boon for performance/efficiency as well. But they require programmer knowledge and intuition. This, along with the lack of generalizability, can be thought of as barriers for wide scale adoption of these languages.