I have been pondering the same concept for years. The problem seems to be: A) Making AI that is able to plan for goals set by players and achieve them is not just "not mainstream", its actually unique when its used. Even the algorithms like GOAP (Goal Oriented Action Planning) is very limited (eg cant really chain goals, one goal at a time,...), and while better algorithms have been proposed (UDGOAP - Utility Driven GOAP), they havent been implemented in any game yet. B) This type of game is not proven, so noone is gonna invest big bucks. And implementing completely new and untested mechanics and core loops will likely burn more cash as you iterate for usable outcome.

Theres a wishing for games of the future, tho