I think they’re saying they changed to referring to Alexia in conversation to A or Her

I'm responding to "devices will be much [better] at figuring out when we're addressing them than the primitive idea of a wake-word". That kind of on-device context detection may not be feasible with the currently-shipping hardware.

The manufacturers want you to buy new hardware at least every year, so that's not a barrier to new features unless you refuse to upgrade.

I'm sure they could do that detection without local updates in the future (disable the output if it detect a wrong context).