Given where this is going, couldn't we just skip a couple iterations of the hype cycle, and simply port BEAM runtime to work directly on machine clusters instead of machine code? This way we'll jump straight to one actor = one microservice = one VM.