For your first point where you are attempting to impose agreement, I believe the other commentator is saying that tradeoffs are non-negligible between the two.

Sounds like the difference between edge and centralized ML scoring.