The Five Scoping Questions
Five questions that draw the boundary between AI's work and yours before any prompting starts — so you don't hand AI the one part of the job only you can do.
The five questions that draw the boundary between AI’s work and yours before any prompting happens — because the most expensive scoping mistake isn’t giving AI too little, it’s giving it the one thing only you can do. An expert who hands AI the central argument and keeps the typing has the division exactly backwards, and won’t find out for months.
The output feeds straight into the Universal Kickoff Prompt as its scope answer. When judgment is on the line, hallucination risk is why the verification step stays yours.
The Five Scoping Questions
Answer in writing. Twenty to thirty minutes.
1. WHAT AM I PRODUCING?
Name the deliverable specifically — not "a market analysis"
but length, contents, audience, and deadline.
2. WHAT'S THE QUALITY BAR?
Name what good enough looks like: "sendable to the client
without embarrassment" / "publishable under my name" /
"good enough to think with."
3. GIVEN THAT BAR, WHAT CAN AI RELIABLY CONTRIBUTE?
Drafting from my structure · transformation between formats ·
extraction · pattern recognition in known domains · code
under supervision · expansion of my points · summarization.
Which of these apply here?
4. WHERE'S THE HUMAN-ONLY CONTRIBUTION?
Novel reasoning · judgment only an expert in this can make ·
anything I can't verify. Which parts of this project are those?
(Be honest — under-claiming this list is the expensive move.)
5. THE BOUNDARY, EXPLICITLY STATED
One or two sentences naming where AI's contribution ends and
mine begins. Write it down. Pin it to the top of the
working doc.
Why it works Implicit boundaries drift. If you haven't decided where AI's contribution ends, then at 11pm under a deadline it quietly expands — the strategic judgment that was supposed to be yours becomes AI's by default, and by the time the project's in trouble, the boundary moved without anyone choosing it. The written statement is cheap insurance against your own future fatigue. And question 4's parenthetical is the whole game: the temptation is always to under-claim the human-only list because giving more to AI feels efficient. The cost of under-claiming is measured in rewrites.
Then what The output is a scope statement that fits on a notecard — feed it straight into the Universal Kickoff Prompt as question one's answer. If writing it reveals the project has no clear deliverable or quality bar, that fuzziness is what's been costing you; it just became visible, which is the fix starting.
These prompts put the framework from Before You Build in your hands. More about the book →
Frequently asked questions
How do I decide what AI should and shouldn't do on a project?
Answer five questions in writing before prompting: what you're producing, the quality bar, what AI can reliably contribute, where the human-only contribution is, and one explicit sentence naming the boundary between AI's work and yours.
What's the most expensive scoping mistake with AI?
Giving AI the one thing only you can do — handing it the central argument or novel reasoning and keeping the typing. Under-claiming the human-only list is the costly move, and it's measured in rewrites.