Ask for options and tradeoffs before you decide
To decide with AI, ask for two or three options with the tradeoff, the cost, and what you might be wrong about. Do not ask what you should do. You choose.
Next step: copy the first prompt that matches your task. Fill in the brackets. Send only that.
A prompt is the message you paste. An approval gate is a stop you control. A checkpoint is a pause after one phase. School pages teach. They do not do the homework. Follow your teacher's rules.
Last updated
How do you do it?
Do the steps in order. Each line is one action, and you stop at the end of that action.
- 1 Skip the question that hands over the choice. "What should I do?" invites one answer and a pep talk. Ask for options instead.
- 2 Ask for two or three options. Each option needs a name, who it is for, and the fact from your notes that it uses. Two or three is enough. A list of ten is a stall.
- 3 Ask for the tradeoff, the cost, and the weak spot. A tradeoff is what you give up. A cost is time, money, or a risk you can name. The weak spot is what you might be wrong about.
- 4 You choose, then name the one next step. Say which option you picked and why, in your words. Then ask for a draft of that option only. The other options wait.
Work
What prompt fits “A choice your boss will ask about”?
You have a real constraint. You want the options before the meeting, and you still want to be the one who decides.
Do not tell me what I should do. I will choose.
Decision: [one sentence, such as whether to move the weekly update to a short doc].
Facts: [the real limits: time, people, deadline, what already failed].
Reader: [who will ask you about this].
Give me 3 options. For each one, include:
- what we would do
- the tradeoff (what we give up)
- the cost in time or money, using only facts I typed. If the cost is not in my notes, write COST UNKNOWN
- what I might be wrong about
Do not rank them. Do not end with a recommendation or encouragement. Then stop.School
What prompt fits “Two ways to study, and you pick”?
You are choosing how to practice. The chat must not pick your thesis or write the assignment.
This is for learning. Do not write anything I will turn in. Follow my teacher's rules.
Do not tell me what I should argue. Do not write anything I could turn in.
Decision: how I will study [topic] before [date].
Facts: I have [minutes]. I already understand [what you know]. I am stuck on [what is hard]. My notes: [a few lines you wrote].
Give me 2 study options. For each, say what I will do in my own words, the tradeoff, and what I might be wrong about if I pick it.
Do not rank them. Then ask me which one I choose. Wait.Life
What prompt fits “A family plan with a real limit”?
You know the budget, the people, and the thing that cannot move. A price is not real until you see it.
Do not tell me what we should do. I will choose.
Decision: [Saturday plan, or how to handle this bill].
Facts: people [who]. Budget or amount [number you typed]. Must keep [the constraint]. Will not [the limit].
A price that is not in my notes is COST UNKNOWN until I see it on the real page.
Give me 3 options. For each, say what we do, the tradeoff, the cost using only my numbers, and what I might be wrong about.
Do not invent hours, fees, or phone numbers. Do not rank them. Then stop.What mistakes should you avoid?
These are the misses that waste the result. Skip them before you send anything.
- Asking "what should I do?" and then doing whatever comes back first.
- Letting the chat rank the list so the first row feels like the decision.
- Accepting a cost, a date, or a risk that was not in your notes.
- Asking for options and then skipping the line about what you might be wrong about.
More in this section: Stop the yes-man reply · Ask for a plan before the work · Approve the plan before it continues
Practice on a real task: Compare two products before you buy · Write prompts that get useful work
Last updated · All director guides
Where does this come from?
These are public pages. The note after each link is what that page says, in plain words.
- Anthropic, “Trustworthy agents in practice” (April 9, 2026). Anthropic says an agent plans, acts, looks at the result, and repeats until the job is done or it needs a person. In Claude Code, Plan Mode shows the plan first. You can edit it and approve it before any work starts, and you can still step in later.
- Anthropic, “Building effective agents” (December 19, 2024). Anthropic says an agent can pause for a person at a checkpoint. One pattern they describe is to write an outline, check that the outline is good, and only then write the piece.
- OpenAI Help, “ChatGPT Work and Codex”. OpenAI describes Work as an agent for longer, multi-step jobs. The help page says you can review progress, change direction, and approve important actions. If your account does not show Work, use the same stops in a normal chat.
What else do people ask?
Open a question. The short answer under it is the one to use.
What should I ask instead of "what should I do?"
Ask for two or three options. For each one, ask for the tradeoff, the cost, and what you might be wrong about. Then you choose.
How many options should I ask for?
Two or three. Enough to see a real difference. A long list is harder to compare and easier to ignore.
Can the chat decide for me?
It can sound like it did. That is the failure. You name the option you are taking, in your words, before it drafts anything.
How is this different from asking it to poke holes?
A hole-poke assumes you already like one idea. Options come first, when you have not chosen. After you choose, you can still ask what is weak about that one.