That's gonna be a while coming - we're now entering a stage where we won't even understand the code that gets written.
Now, sure, some holdouts will understand the code, but that's not going to be the norm soon.
But I do wonder if LLM-driven code analysis might actually increase code comprehension and agency for laymen?
I've been quite impressed with AI's ability to decipher and visualise code and system relationships in e.g. mermaid diagrams.
Perhaps the representation of code will become more elastic (i.e. you have literal code, or AI-produced translations you can directly manipulate)
Coming from HomeKit, I've tried (_really_ tried) to move to Home Assistant. The gap in usability was too enormous for me to cross, and I'd brand myself a hacker.
I won't trust any kind of a "smart" device to operate the front door lock - ever - but smart lightbulbs are still stupid lightbulbs. I can just flick the switch.
Privacy concerns are valid - I can be profiled based on usage. But it's not like Apple doesn't know my precise location already.
With all that in mind, I'd say usability comes first.
If the lock freaks out - and yeah, I'm having problems with my HomeKit stuff every now and then - I need the physical key on me. That already defeats the purpose. Otherwise I need to grab the spare key from wherever I keep it. This is not a mere inconvenience, this is an emergency procedure.