Planning is the model breaking a goal into ordered steps.
Explicit planning helps a lot on tasks with several dependent stages.
ReAct interleaves reasoning and acting, one step at a time.
Plan-and-execute writes the full plan first, then works through it.
Plans need to be revisable, since the first step often reveals the plan was wrong.
Planning is a model breaking a goal into ordered steps before acting.
Without it an agent tends to act on the first plausible move and drift.
With several dependent stages, thinking first measurably improves the outcome.
Plan-and-execute writes the whole plan up front and then works through it.
That is efficient and legible, and it commits before anything has been observed.
ReAct interleaves instead, reasoning one step at a time between actions.
Each observation informs the next decision, which handles surprises better.
It costs more model calls and produces a less tidy record.
The practical answer is usually a rough plan up front that is allowed to change.
The first step reveals what the plan got wrong more often than not.
So a plan must be revisable, and a rigid one is worse than none.
Ask for the plan explicitly and show it to the user before anything runs.
That single habit catches most of the badly conceived attempts.
Long plans degrade, since a ten-step plan usually breaks by step four.
Break large goals into smaller runs rather than one heroic sequence.
One question on this card. Every option explains itself once you pick, right or wrong.
17 concepts in this module, then a graded test with every answer explained.