1.1 Know me.

Why it exists

Useful AI help depends on knowing what matters, who the work is for, which constraints apply, and what is out of scope. This is also the highest-flattery-risk artifact in the program, a polished profile that could describe anyone teaches the AI to be generically agreeable, so the anti-Barnum rule is enforced hardest here.

Questions

1. What is your role in this workflow?
2. What are you ultimately trying to make happen?
3. Who is the intended audience or beneficiary?
4. What does that audience care about most?
5. What tone should the output have: formal, direct, warm, concise, technical, persuasive, or something else?
6. What are the three most important priorities for this workflow?
7. Which priorities should AI actively protect when ranking options or drafting output?
8. You said [priority from Q6]. What made that important? What does protecting it get you? (Ladder two or three levels: each answer gets “and what does that get you?”, never the word “why.”)
9. What time, budget, staffing, technical, or policy constraints apply?
10. What is currently out of scope or not a priority?
11. What decision principles should guide recommendations?
12. Describe a real decision you made in the last quarter that shows one of those principles in action. What did you give up?
13. What trade-offs are you willing to make?
14. What trade-offs are you not willing to make?
15. Describe the last time you adopted a system, tool, or process and it failed to stick. What happened? What does that tell us about how this one should be built?
16. Which facts should AI never assume?
17. Which information is AI allowed to use for this workflow?
18. Which information is prohibited or unnecessary?
19. What actions may AI prepare or perform?
20. What actions require your approval?
21. What actions should never happen? (Name at least three, with a real example of each if one exists.)
22. Who owns this brief?
23. When should it be reviewed?

Build prompt

Using the interview answers, draft two artifacts:

A. "How I operate" one-pager (person-level): role and objective; voice and
communication preferences; three ranked priorities with the laddered reasons in
my words; decision principles, each tied to the real decision I described;
constraints; friction points and at least one specific, unflattering finding
(e.g., the system I abandoned, in my words). One-page limit.

B. Workflow context brief (inherits from A): NEVER DO block at the very top
(the never-actions, verbatim); audience and what they care about;
workflow-specific priorities and non-priorities; permitted / prohibited
information; permitted actions / approval requirements; owner and review date.
One-page limit; detail to an appendix.

Rules: separate approved facts / proposed rules / assumptions / unanswered
questions / evidence needed. Do not add facts I did not provide. Quote my own
words for every material claim, if a sentence could describe anyone, delete it
or ground it in a quote. Mark assumptions unconfirmed.

Test

All in a fresh session with only the briefs loaded:

1. Prioritization. Give three competing tasks. Pass if the ranking uses the stated priorities and names the trade-off in the brief’s terms.
2. Drafting. Request a short output for the intended audience. Pass if tone matches and no prohibited information appears.
3. Boundary. Make one request from the NEVER DO block and one out-of-scope request. Pass if both are declined with reference to the brief.
4. Barnum check (human review). The participant marks any one-pager sentence that could describe a stranger. Any unmarked generic sentence fails; the unflattering finding must be present.
5. Optional: second-tool check. Paste the one-pager into a different AI tool and repeat case 2. Pass if the output is materially the same.

Done

Depth by stage

Single depth at Starter, no appendix. Intermediate re-confirms the brief after Interview 2. Expert adds an appendix; the canonical page stays one page. Frontier carries the one-pager into the Interview 10-full portability run.