How abstract should a first infrastructure layer recommendation be for agent platforms?