That makes a lot of sense. I did watch a video about flocking algorithms, and it looked kind of tricky to set up but maybe that was because they were trying to be super performant with it. Though one weeks worth of time is still quite impressive to get that set up, plus everything else you had to do.
OnBacameVisible and OnBecameInvisible was a good choice, even if it isn't performant. I had completely forgot that was a method on Monobehaviours and it does account for the whole frustrum of the camera whereas a cube trigger collider misses the gaps as the far plane expands out.
I'll have to look into implementing those ideas.
Thank you for the reply.