I'm thinking of an assistant where I show (or tell it) what I want, and it just does it (it 'learns'), improving based on feedback from the users.
Hard-coding times in the day, for example, could be improved if the assistant knew what would be the ideal time to sleep to me, based on all the data collected for each profile of the people living in that home. It might ask me or suggest me things to do, having more flexibility than static rules.