A new approach in multi-agent reinforcement learning, called Diamond Attention, introduces structured randomness to improve coordination among identical agents, enabling better role differentiation and task performance. This method outperforms traditional deterministic policies in various scenarios, highlighting the importance of randomness in achieving effective agent collaboration.