Practice: Planning: breaking a goal into steps
Self-check
Section titled “Self-check”Seven short questions. Answer each in your head before opening the collapsible. Active retrieval is where the learning sticks.
1. In one sentence, what is the planning turn?
Show answer
It is the move from deciding one action at a time (reacting) to deciding the shape of the whole job first: the agent breaks a goal into an ordered set of sub-tasks before it starts executing.
2. Planning is which one core operation?
Show answer
Decomposition. Taking one large goal and breaking it into an ordered list of smaller sub-tasks, each achievable in a step or a few, where the order captures the dependencies between them.
3. Name the three ways a purely reactive agent fails on a long task.
Show answer
It does things in the wrong order (it cannot see dependencies), it repeats or skips work (it has no map of what is done), and it drifts off the goal (nothing holds the overall shape in view). A plan is the map that fixes all three.
4. Why is producing a plan described as a “reasoning move”?
Show answer
Because it is the same skill as a model “thinking step by step”: instead of asking the model for the final answer, you ask it first to lay out the intermediate steps that get there. Laying out the steps before acting is reasoning applied to the structure of the task.
5. What is the difference between plan-then-execute and replanning?
Show answer
Plan-then-execute builds the whole plan upfront and runs it, which works when the task is predictable. Replanning revises the remaining plan as results come in: the agent executes a step, observes the result, and adjusts the rest of the plan when reality contradicts it. Most capable agents sit between the two.
6. Why is “arrange the whole trip” a bad first step in a plan?
Show answer
The grain is too big. It hides its own complexity; it is the original goal wearing a list’s clothing, not a plan. A useful step is a single clear intention the agent can act on in a turn or a few. (The opposite error is steps so small the plan reads like a transcript of keystrokes.)
7. Someone says, “Our agent produced a detailed plan, so the task is basically done.” What is wrong with that?
Show answer
A plan is the map, not the journey. Producing a beautiful plan is not progress until the steps are actually executed and checked. Worse, a plan built before any action is a hypothesis, not a contract; if results contradict it, the agent should replan, not treat the plan as finished work.
Try it yourself: decompose a goal at the right grain
Section titled “Try it yourself: decompose a goal at the right grain”No tooling, no cost; this is design judgment. Write a 4-to-6 step plan for the goal below, the way the lesson decomposed the Tokyo trip. Aim for the useful grain: each step a single clear intention, and the order capturing the dependencies. Then check your steps against the notes.
Goal: “Onboard a new hire: get them a laptop, accounts, and a first-week schedule.”
Show a sample plan
1. Confirm the hire's role and start date (drives everything else)2. Order the laptop and required equipment for that role3. Create accounts (email, internal tools) for the start date4. Build a first-week schedule (orientation, team intros, training)5. Send the schedule and account details ahead of day oneYours does not need to match. What matters: step 1 establishes the facts the later steps depend on (you cannot create role-specific accounts before you know the role); each step is one clear intention, not a keystroke transcript and not a whole project; and reading the list back tells you the strategy at a glance. If a step still feels like a project (“set everything up”), break it down; if it reads like clicks (“open the HR portal, click new hire”), pull it up a level.
Try it yourself: where does replanning kick in?
Section titled “Try it yourself: where does replanning kick in?”For the plan below, the result of one step contradicts the plan. Identify which step forces a replan, and write the revised step, the way the lesson did for the over-budget flights.
GOAL: ship a feature this sprint. 1. estimate the work -> 8 days of effort 2. assign it to one engineer -> only 4 working days left in the sprint 3. build it 4. test and releaseShow answer
Step 2 forces the replan: 8 days of effort will not fit in 4 working days with one engineer, so the plan no longer fits reality.
REPLAN at step 2: effort (8d) exceeds remaining capacity (4d, one engineer). -> revised step 2: split the work across two engineers, OR cut scope to a 4-day slice and defer the rest, then continue the plan from there.A rigid plan-then-execute agent would have plowed ahead and missed the sprint. A replanning agent notices the result contradicts the plan at step 2 and revises the remaining steps before building. That is the same self-correction instinct as retrying a weak retrieval, lifted to the level of the whole task.
Flashcards
Section titled “Flashcards”Ten cards. Click any card to reveal the answer. Use the Print flashcards button to lay out the full set as one card per page for offline review.
Q. What is the planning turn, in one sentence?
The move from reacting one action at a time to deciding the shape of the whole job first: the agent breaks a goal into ordered sub-tasks before executing.
Q. Planning is fundamentally which operation?
Decomposition: breaking one large goal into an ordered list of smaller sub-tasks, where the order captures the dependencies between them.
Q. What three failures does a purely reactive agent hit on a long task?
Wrong order (cannot see dependencies), repeated or skipped work (no map of what is done), and drift off the goal (nothing holds the overall shape).
Q. Why is producing a plan a 'reasoning move'?
It is the same skill as thinking step by step: you ask the model for the intermediate steps first, rather than jumping straight to the final answer.
Q. In what form is a plan usually written, and why?
A structured, executable form (a numbered list or machine-readable steps), not loose prose, so each step can be followed, checked off, and executed one at a time.
Q. What is plan-then-execute?
Build the whole plan upfront, then run it in order. It works when the task is predictable.
Q. What is replanning?
Executing a step, observing the result, and revising the remaining plan when the result contradicts it. The same self-correction instinct as retrying a weak retrieval, at the level of the whole task.
Q. What is the right grain size for a plan step?
A single clear intention the agent can act on in a turn or a few. Too big and it hides complexity (the goal in disguise); too small and the plan reads like a keystroke transcript.
Q. When is planning NOT worth it?
On single-step tasks (one lookup, one transform). The extra reasoning step and rigidity are pure overhead. Planning earns its keep on multi-step, interdependent goals.
Q. Is a finished plan the same as finished work?
No. The plan is the map, not the journey. It is not progress until the steps are executed and checked, and it should be revised if results contradict it.