I think you're on the right track, software tries to layer itself as much as possible. New architectures and capabilities will only make the concepts and tools exposed by the glue between the layers change.

Unless the new thing isn't Turing-complete and can't be implemented with a Turing-complete system, it will be abstracted away at first just so we have an environment to start building with, and can start using it without reinventing every single wheel we have.