Coming up with a more useful Indonesian UP is on the low prio part of my to do list, I will keep this idea in mind for it. I think you raise three mostly independent action items here: - Stop the AI from pillaging independents: that makes sense all around and is easily implemented - Reduce overall AI propensity to pillage: might be less easy to change, but I am not familiar with the related code at all. Maybe there is one part of the AI that is responsible for the indiscriminate pillaging that I can limit/disable. - Stop the AI from constant improvement replacement: I am really not sure what is causing this because I did not touch the code besides adding values for new improvement effects (such as health/happiness), which shouldn't have an impact on the replace behaviour. I encountered this "bug" before and it's really hard to track down what is responsible for it. In general the AI should be able to decide that it wants a different improvement that is regarded as more useful, but I feel like it should be less willing to replace a grown improvement, and apparently the algorithm has issues where the currently present improvement always is worse than some other leading to constant replacement. I tried before to put an end to that but without much success. Maybe the only or at least better way to go here is to start using the K-Mod improvement AI.