Agent Partitioning with Reward/Utility-Based Impact