I would also just add to this that actor components are pretty finicky when it comes to keeping them any where out of the actors themselves. Moving them from actor to actor also costs performance.
This thread also tackles the problem of if you are driving logic in actor components with tick enabled inside of them you will pay the penalty of a single actor constantly evaluating multiple ticks.
I would make a uobject (object) derived class that would get it’s tick update from a central ai manager. Guess it all depends on how complex you want the logic inside of the ai building blocks.
Personally I’d probably start off with state trees as their logic seems a bit more straightforward than behavior trees and already look like an ok system. They are internally optimized.