I hate to see its evolution as never-quite-there toward a never-quite-decided future.
Whatever they do, the years-long release cycle needs to die, allow whatever improvements you'd like on head without worrying much about BC, occasionally cut off a LTS release that never gets anything new without 100% BC. Maintain these long enough that it's economically feasible for people to base projects on (Whatever the benefit of the project is * years that benefit is durable > cost of development). Focus on data migration for when people re-build sites on newer platform instead of continuous upgrades -- nobody does these unless they're painless, they will never be painless.
There's too much coupling between presentation, client-side behavior, server-side behavior and data model for me to see this particular change going anywhere good.
You can't truly decouple client and server while still letting non-developers build web apps unless you build a UI-configurable framework for consuming an API generically and building client-side presentation and behavior. Even then, unless you also go so far as to infer things about my data from its API, it's still twice as much work to describe both my server-side domain model and its presentation instead of letting one be implied by the other.
But in typical Drupal fashion, they probably don't mean "decouple" when they say that word, so it's really just rewriting the current hodge-podgy fully-coupled client-side behaviors in a more structured but still coupled way using a client-side framework -- absorbing it the way they did Symfony for the server, not the best move but definitely with value and very achievable.