Using Recursive Prompter

# Using Recursive Prompter — three worked examples

Three compressed illustrations of what the interview looks like in practice. Real transcripts would run 30+ exchanges; these are sketches.

## Example 1 — Redesigning a weekly workflow

**You paste the prompt, then say:** "Interview me about my Friday afternoon marking routine. Something isn't working but I'm not sure what."

**Phase 1 — Behaviour.** The AI asks you to walk through what actually happens on Friday at 3pm. Not the aspiration. You describe opening the stack, opening the laptop, reading the first essay, getting a text, getting distracted, closing the stack an hour later having marked two. The AI pushes on specifics: last time this happened, where were you sitting? When did the text come in? How long did you actually read before distraction?

**Phase 2 — Friction.** "Where does this break?" You say the distraction is the thing. The AI pushes back: is it? Or is the distraction welcome because the marking itself isn't moving? You go quiet. You admit the essays feel undifferentiated — all the same two complaints, you've written the same comments a hundred times.

**Phase 3 — Hidden Assumptions.** "What are you treating as fixed?" The AI asks. You say the comments need to be personalised. It asks: do they, really? Or could the common comments be a separate artefact the students all get, and your personalised time goes only to what's individual? You sit with that for a minute. You hadn't considered it.

**Phase 4 — Redesign Criteria.** The AI asks what "better" looks like. You say: I want to spend my thinking time on what's individual, not what's common. I want the common feedback captured once, reused. I want to finish the stack in half the time without lower quality.

**Exit.** You leave with a named bottleneck (undifferentiated comment writing) and a concrete next move (draft a one-page "common comments" artefact to attach to the whole batch, leave only the individual lines for each essay).

---

## Example 2 — A fuzzy project idea

**You paste the prompt, then say:** "I have an idea for a tool that helps teachers turn rubrics into feedback comments. Interview me — I don't know the shape of it yet."

**Phase 1 — Behaviour.** The AI asks you to describe the current way teachers get from rubric to comment. You talk through it. It pushes: who specifically? What subject? What kind of rubric? You realise your imagined user is a TOK teacher marking an essay, not a maths teacher marking a problem set, and those are different problems.

**Phase 2 — Friction.** Where does the current process break? You say it's slow and the comments drift in quality over a stack. The AI asks what "drift" means. You admit: by essay 30 your comments are shorter and less specific than on essay 1. That's the friction.

**Phase 3 — Hidden Assumptions.** What are you treating as fixed? The AI asks. You say the comments need to be fully written by the teacher. It asks: do they? Or could the AI generate a draft from the rubric criteria and the teacher edit? You realise you'd been assuming AI-draft was somehow not legitimate — a hidden norm, not a real requirement.

**Phase 4 — Redesign Criteria.** What does success look like? You say: a teacher uploads the rubric, uploads the essay, gets a draft comment in 10 seconds, edits for voice and specifics, pastes into the report. Quality doesn't drift because the AI doesn't tire.

**Exit.** You have a named project shape you didn't have going in. You now know: the tool is for TOK essays, the job is draft-then-edit, the hidden assumption was about legitimacy. Run Work Structurer next.

---

## Example 3 — Diagnostic on a stuck creative project

**You paste the prompt, then say:** "I've been stuck on a blog post for three weeks. Interview me about it."

**Phase 1 — Behaviour.** The AI asks what happens when you sit down to write. You describe opening the draft, reading what's there, adjusting a sentence, closing it again. The AI pushes: how much time, how often? You realise it's five minutes a day, twenty times, and you've barely added a word.

**Phase 2 — Friction.** Where does it break? You say the opening doesn't work but you can't see why. The AI asks if the problem is the opening or something earlier — the premise. You pause. You say maybe the premise.

**Phase 3 — Hidden Assumptions.** What are you treating as fixed? The AI lists what you've said — the title, the angle, the structure. It asks which of those you've actually committed to. You realise the title came from a half-remembered idea three weeks ago and you've been defending it without ever testing it.

**Phase 4 — Redesign Criteria.** What does "done" look like? You say: a reader finishes it and forwards it to someone else. The AI asks what belief in the current draft would make that happen. You can't name one. That's the finding.

**Exit.** The post wasn't broken — the premise was. You have a next move: rewrite the premise from scratch, test it in one paragraph, decide whether to continue or kill the post.

---

## The pattern across all three

- The first answer is rarely the real answer.
- The AI's job is to push past the aspirational framing.
- The exit is a named bottleneck and a specific next move, not a solution.
- You finish slightly tired. That's the signal.