Automated planning is the branch of artificial intelligence concerned with choosing actions to achieve a goal. Given a description of the world’s current state, a set of available actions and what each one changes, and a goal to reach, a planner computes a sequence of actions that gets from here to there. It is the part of AI that answers “what should I do, and in what order?” rather than “what is true?” or “what category is this?”
The classic formulation comes from the STRIPS planner that Richard Fikes and Nils Nilsson built at Stanford Research Institute in 1971 to control the Shakey robot; the STRIPS paper is cited here as the field’s foundational document. In this model the world is a set of facts, each action has preconditions and a list of facts it adds and deletes, and planning is a search through possible world states. Because that search can explode combinatorially, much of the field is about taming it: drawing on the General Problem Solver’s means-ends analysis, using informed search like A*, and computing heuristics automatically from the problem description. Since 1998 the International Planning Competition has driven progress by pitting planners against shared benchmarks written in the PDDL language.
Planning sits between two extremes. Pure logic can be too slow; pure learned policies can be hard to guarantee. Classical planners offer transparent, goal-directed reasoning with clear correctness, which is why they remain in use even as learned methods spread.
Why business readers should care: automated planning is the engine behind production scheduling, vehicle and warehouse routing, spacecraft operations, and the step-by-step reasoning increasingly asked of AI agents. It turns a stated goal into an executable sequence of actions.