Planning

Planning is an AI Agent's ability to design, order and — when necessary — revise the sequence of steps it takes to reach a goal. Classical AI had its STRIPS and PDDL planners; in the LLM era the dominant pattern has become 'produce an explicit plan, then execute it step by step', a recipe that makes long chains noticeably more reliable. Modern agent systems often delegate planning to a Reasoning model or a dedicated 'planner' role, with execution handled by a faster, cheaper 'executor'. A good plan is half the work; a bad one quickly exhausts whatever Reflection Pattern and Self-Critique you bolt on top.