This is, for the time being, mostly an experiment. I was able to get consistent results where a diverse community of agents could provide a more accurate answers for tasks that they would fail if asked to perform individually.