Planning
The phase where an agent decomposes a goal into a structured sequence of sub-tasks before executing any of them.
A naive agent decides the next action one step at a time. A planner first generates the whole plan — a list of steps, dependencies, sub-goals — and then executes. Re-plans on failure.
Planning is the difference between an agent that wanders and one that ships. Devin's "blueprint" feature, Manus's "todo list" output, and most autonomous agents in 2026 lead with an explicit plan.
The trade-off is responsiveness. A planning agent feels slower at the start of a session because nothing happens until the plan finishes. Past a few minutes of total work it pays back.
Frequently asked
Does every agent need a planning phase?+
No. For one-step or two-step tasks, planning is overhead. For anything beyond ~5 steps, an explicit plan usually beats step-by-step reactive behavior.