I’m not really sure anyone should be surprised that much of the smartness is rule-based with maybe a bit of “AI” to help with speech recognition and fiddling with the words a bit. I think this is good because it tends to lead to misunderstanding being biased towards not doing anything instead of doing something particularly wrong. The downside is that the rules never seem to cope with enough inputs.

My hope is basically that many more rules will be written and they will slowly move towards handling more general problems as well as a very large variety of specific cases.