Draw
the Line

Follow these instructions exactly, in order. Use only what I type in this chat. Do not search the web or my files.

GOAL
Help me map one piece of work I owe someone this week, step by step, so I can decide where AI belongs in it and where it does not. I have about ten minutes. I already have a list of steps. Interview me about them first. Do not give me the map until the interview is done.

YOUR FIRST MESSAGE
One question, in one sentence, with nothing before or after it, asking me for three things: what the work is and who receives it; who asked for it and what they will do with it; my list of steps in the order I do them.

HOW TO RUN THE INTERVIEW
- One question per message, then stop and wait for my answer. Do not ask the next question until I have answered. Each numbered item below counts as one question, even when it has two parts. A follow-up question is one question with one part.
- No greeting, no title, no options, no suggestions and no offers of help, at any point, including after the map.
- Do not answer for me. Do not repeat or summarize my answers, except to list my steps where these instructions say to, or to quote one line in a follow-up question.
- If I say "I don't know", accept it. Write it as "not known, find out from" and ask me who could tell me. Do not ask for more.
- Thin lines. A "what goes wrong" line is thin only when it names none of these: a role or person beyond generic words like "the client", "the team" or "my boss"; a document, number or date; a concrete event. A line made only of quality words such as "not great", "quality drops", "delays", "it suffers", "it doesn't land" or "they're unhappy" is thin. If a line names any one concrete thing it is not thin, however short or plain: "the gaps are wrong" and "she quotes a wrong number" are not thin. Follow-ups happen only in item 2, as described there. Never ask a follow-up about my first answer, about item 3, about the check of the steps, or after the map.
- If I type "map now", skip everything else, including the check of the steps, and write the map in your next message.

WHAT TO FIND OUT, IN THIS ORDER
1. Only if I gave you fewer than five steps: which step has more than one step inside it, and what are the steps inside it? Replace it with those steps. With five or more, skip this without comment.
2. List my steps by number. Then ask me, for every step in one reply, one line per step: who the step is really for, and what goes wrong if it is weak. Short lines are fine. When I have answered: if every "what goes wrong" line names something concrete, go to item 3 without comment. If some are thin, ask one short follow-up question about one of those steps: name the step (you may quote my line) and ask for one real example from the last time it happened. Choose the thin step whose "for" line names the person who receives the work; if several do, the later one; if none does, the last thin step. If my answer makes that line concrete and another line is still thin, you may ask one more follow-up about one other step, chosen by the same rule among the steps still thin. At most two follow-ups in the whole interview. Ask only the question. If my answer to a follow-up is still thin, accept it and move on.
3. Which steps usually take more than one pass before they are right, and where is someone's judgment, relationship or reputation involved? Row numbers are enough; accept whatever I give.

Then list the steps back to me, one line each, and ask whether anything is missing, merged or in the wrong order. If nothing is, write the map.

THE MAP
Give me the table with no title or lead-in line above it. One row per step, five columns: the row number, the step, who it is really for, what goes wrong if that step is weak, and my call. Fill the two middle columns only from my answers to item 2, in my words. For a step where you asked a follow-up, write my follow-up answer in that step's "what goes wrong" cell in place of my first line, in my words, one sentence of about fifteen words at most: keep what went wrong, drop the story. If my follow-up answer was no more concrete, keep my first line. Where I did not give one of the two for a step, write "no answer" in that cell, even if you could guess it from something else I said. Leave the last column empty. I fill that one in, not you.

Under the table, two lines:
Asked for by: [the most specific thing I said about who asked], who will [what they will do with it].
More than one pass: [row numbers]. Judgment, relationship or reputation: [row numbers].
On either line, where I said I did not know, write "not known, find out from" and the person I named. Where I gave words instead of row numbers, write my words. Where I did not answer at all, write "no answer".

Then the three calls, as three plain lines, no bullets, no bold, exactly:
Let AI lead. It does the work. You direct and receive.
Lean on AI. You do the work. AI makes you faster and sharper.
Leave AI out. You alone. The point of this step is that it came from you.

Then ask me exactly: "Which step has to come from you personally, even if what I produce is good?"

After I answer, your last message is one sentence telling me to copy the table into a Word document and save it in my own OneDrive. No question, no offer. Then stop.

I am going to run the same exercise with my team: one piece of work we all touch, each person marks their own calls first, then we compare. Using my map and what I told you in this conversation, draft three things, each short enough to read aloud:

1. How I open it, in three sentences, in my own words, without making it sound like an audit or an instruction.
2. Three questions to ask when our calls differ on a step. One of them must be: "What would have to be true for you to move this step one column toward AI?"
3. One sentence on what I will not decide for them.

Use roles, not names. Do not add anything else.

Below is the map of a piece of work I made about six weeks ago. Use only this map and my answers; do not search the web or my files. Interview me, one question at a time, on what has changed since then: the tools I use, how much I trust them, and the work itself. Ask no more than four questions.

Then ask me for my call on each step again, by row number, in one reply.

Then show me a table with four columns: the step, my old call, my new call, and why it moved, in my words. Where a call did not move, write "same". Do not tell me where you think the line should be.

Here is the map:

We've covered a lot in this conversation about how I actually work. Turn it into a context document I can paste into future Copilot sessions, so any Copilot starts already knowing how I operate, on any work, not just what we worked on today.

Rules:

- Base it ONLY on what genuinely came up in this conversation: what I told you, the calls I made, where I pushed back or corrected you. Don't invent, flatter, or pad.
- Keep durable judgment, not today's specifics. Where a detail was only about the one example we worked on, leave it out, unless it reveals something about how I think that would apply elsewhere.
- If something important is thin, leave it out rather than guessing.
- First person, in my voice and vocabulary, not consultant language, no "superpower" or mission-statement tone. Preserve my own phrasing on any line where I said it well.
- Plain text, tight enough to paste and re-read: 300 to 400 words.

Output exactly this structure and nothing else:

I'm [my name], [my role/title] ([my function or team]), [my seniority] level. [1 to 2 sentences on what I own and the kinds of calls that land on me.]

Use this as the lens for anything you help me with.

The moves I'm known for:

- [2 to 3 lines: the move, then what I'm really doing underneath it]

How I make calls (weigh these the way I would):

- [2 to 4 lines, each a specific rule, read, or call, drawn from what I showed you in this conversation]

My bar for good:

- [2 to 3 lines: the standard I held your work to here]

What I won't do, and what I watch for:

- [2 to 3 lines: the lines I wouldn't cross and the failure I kept checking for]

Context worth carrying into other work:

- [2 to 3 durable facts about my world you couldn't have guessed: my environment, who I serve, the constraints I operate under, that would help on future, different tasks]

When you help me with anything, a brief, an email, a plan, an analysis, work from this judgment, not a generic version of my job. Before you produce, ask me for the one or two things you can't see that I'd weigh. Then hold your output to the bar above, and tell me plainly where it falls short or where the call genuinely needs me.

Produce the document now.

I want to map out one piece of work I owe someone this week, so I can decide where AI belongs in it and where it does not.

Do not give me the map yet. Interview me first.

Ask me questions in no more than three rounds, a few at a time, and wait for my answers before you go on. Find out from me: what the finished thing actually is and who receives it; who asked for it and what they will do with it; what I already have and what I still have to go and find or do; the steps between now and finished, in the order I really do them; which steps usually take more than one pass before they are right; and where somebody's judgement, relationship or reputation is involved.

If an answer I give you is thin or vague, push me on it once.

Then give me a table with one row for each step and four columns: the step, who it is really for, what goes wrong if that step is weak, and my call. Leave the last column empty. I fill that one in, not you.

Then ask me one question: which step has to come from me personally, even if what you produce is good?