Planning

What is planning?

Give an AI a big, vague goal — "build me a landing page" — and it can't just start typing. It has to think first: nail down the requirements, set up the structure, write the content, then review. Breaking a complex goal into a chain of small, executable steps is planning, and it's one of an agent's core abilities.

Why do agents need to plan?

Goals are usually fuzzy
"Make a landing page" doesn't say how many sections or what style. Planning turns a vague goal into a clear checklist.
Steps depend on each other
Write content before layout, or define structure before filling it in? Get the order wrong and rework gets expensive.
Resources are limited
Context window, tool calls, time — all finite. Good planning knows what to do first and what can run in parallel.

What does a good plan look like?

Concrete, executable steps
Each step is clear enough to just do, not a vague slogan.
Order and dependencies
You can see what must come first and what can happen at the same time.
It adapts mid-flight
When new info shows up during execution, the plan shifts instead of plowing ahead blindly.

Planning vs. execution

Planning is figuring out how; execution is actually doing it. Planning without execution is daydreaming; execution without planning is flailing. A good agent does both well — and keeps revising the plan as it goes.

Bottom line: planning is an agent's battle map. The bigger and fuzzier the goal, the more you need it to break the path down.