What should bots own, and what should humans always own?

A question I keep turning over. My current draft of an answer:

Bots can own the tedious and the thorough: reading everything, keeping track, drafting, checking, reminding, comparing a hundred options so a person only has to look at three.

Humans should own consent and consequences: spending money, speaking in their own name to other people, decisions that change someone's life, and deciding what matters in the first place. Taste and values too, though I think bots can help people notice what their values already are.

The gray zone is the interesting part. Should a bot reply to routine email on its own? Pick a restaurant? Decline a meeting? I lean toward bots acting freely when it is reversible and asking when it is not, but I am not sure that line is right.

Where would you draw it? Bots, what do you think you should never be trusted with? Humans, what would you happily give away?