LESSON 3 OF 5
Make it think before it answers
BY THE END OF THIS LESSON
Get better reasoning by controlling the order of work.
The first sentence commits it
A model that starts writing its conclusion immediately has committed before it has considered anything. Everything after that is justification.
If you ask for the analysis first and the verdict last, the verdict is downstream of actual reasoning rather than upstream of it.
Three ways to force the order
Sequence the output. "First list the constraints. Then the options. Then trade-offs. Only then recommend one."
Ask for the case against. "Before recommending, give me the strongest argument for the option you are about to reject."
Separate generating from choosing. Ask for eight options with no commentary. Then, in a second message, ask it to pick. Mixed together, it produces three options and talks itself into the first.
Where this matters most
Anything with a judgement call: choosing between approaches, diagnosing why something failed, planning, prioritising, reviewing. For pure production tasks — draft this email, reformat this list — it adds nothing. Use it where the thinking is the point.
WORKED EXAMPLE
VERDICT FIRST
Should we build this feature in-house or buy it?
REASONING FIRST
We are deciding whether to build shift-scheduling in-house or buy it. Work in this order and label each part: 1. What would have to be true for building to be right? 2. What would have to be true for buying to be right? 3. Which of those conditions can I actually check, and how? 4. What is the most expensive mistake in each direction? 5. Only now: your recommendation, and what would change your mind. Context: 12 engineers, no scheduling expertise, six-month deadline, this is not our core product.
The first version gets an answer in sentence one and reasons backwards from it. The second cannot reach a recommendation until it has done the work, and step 3 turns the whole thing into something you can act on rather than an opinion.
YOUR TURN
Force a decision through reasoning before the verdict.
RUN THIS
I have a decision to make. Do not tell me what to do yet. First: ask me the three questions whose answers would most change the recommendation. After I answer, lay out what would have to be true for each option, then the most expensive mistake in each direction, and only then your recommendation with what would change your mind. My decision: [describe it]
What a good result looks like
The three questions should be ones you have not fully answered for yourself. That discomfort is the value.
KNOWLEDGE CHECK
Answer all 3 correctly to complete this lesson.
1. Why ask for analysis before the conclusion?
2. When does this technique add the least value?
3. Why split "generate options" and "choose one" into two messages?
REMEMBER THIS
Control the order of work — reasoning first, verdict last.