Should the AI interview you before you prompt it?

I no longer look for the perfect prompt. I ask the agent what it still needs from me. Based on what it already knows about me and my goals, it finds the gap faster than I would. The answers go straight into the second brain, and only then does the actual task begin. I picked up the base idea from someone else: dump everything about yourself, your career and your goals into the agent, a plain brain dump. That holds up once the second brain already carries substance. While it is still thin, I would rather get interviewed, so the questions guide the dump. The agent then already knows the direction before I start typing. The technique is called reverse prompting. In April 2026 it sat on one of my lecture slides at FH Salzburg, next to self-consistency, reflexion and plan like a graph. The same slide holds the sequence in five steps: talk at length, let it ask you questions, answer, let it hand you the result, iterate. In a school keynote that same month I showed it live, under the line that the AI questions the user before it builds.

Where the idea comes from

I picked up the base idea from someone else: dump everything about yourself, your career and your goals into the agent, a plain brain dump. That holds up once the second brain already carries substance. While it is still thin, I would rather get interviewed, so the questions guide the dump. The agent then already knows the direction before I start typing.

The technique is called reverse prompting. In April 2026 it sat on one of my lecture slides at FH Salzburg, next to self-consistency, reflexion and plan like a graph. The same slide holds the sequence in five steps: talk at length, let it ask you questions, answer, let it hand you the result, iterate. In a school keynote that same month I showed it live, under the line that the AI questions the user before it builds.

The sequence

Four steps, always in the same order.

  • Give context: the agent gets read access to whatever is already written about me, my projects and my goals.
  • Let it interview you: I ask it to question me instead of dictating a finished task. When time is short I set a ceiling, say twelve questions at most, and then it goes straight to the result.
  • Keep the answers: everything that comes out of that conversation goes into the second brain immediately, before the next question arrives.
  • Let it work: only then comes the task. The agent works with the context it just gathered for itself.

The prompt I use

In the "Agenten-Mindset" Deep Dive on 3 August 2026 I show this live. The prompt is one sentence.

Based on what you know about me and my goals, what more information can I provide to you so you can help me better?

For me that one sentence exposed four gaps: a missing hierarchy of priorities between my projects, unclear limits of delegation, unstated goals, and open decisions I had been pushing ahead of me. The agent got a better working environment out of it. The model behind it stayed the same.

I barely teach prompt engineering any more. I say: interview me so I can do X as well as possible. The same line holds for a product, for a text and for a decision. For a prototype it runs longer: interview me first, challenge weak assumptions, niche the idea down, compare alternatives, and only then create. For the technical build a shorter version is enough: ask me the most important questions first, and only ask things that genuinely change the product.

What happens to the answers

Only then do I ask the second question: what task can you take off my plate right now that brings me closer to this goal. The agent answers with something concrete. The suggestions come from what it knows about me and my goals by then, rather than from general advice.

For that to last beyond one evening, the answers need a fixed place. A second brain is an ordered collection of your own notes, decisions and documents, readable by humans and by machines alike. The concept behind it comes from Andrej Karpathy and his note on the llm-wiki, the structure of mine from the Obsidian skills by Steph Ango. Without that place, every interview is gone again the next day.

What the interview needs first

Without existing context the question does little. That is why my prompt block for a fresh second brain opens with a different question: ask me first which folders and data sources you may read, and which stay excluded. Access first, interview second. Why that order matters is in access beats the prompt.

Whoever gets interviewed should also know who is asking. Some models think long and therefore ask sharper questions, others work fast and broad. How I split those roles is in a team of models. You can watch it live at the KI DeepDive, every first Monday, free: the next dates.

Shown live

KI DeepDive "Agenten-Mindset", 3 August 2026, shown live. Prompt wording from the session resources.

  • The first question: what do you still need from me so you can help me faster, based on what you already know.
  • The agent asks live for a real order of priority between the projects, then for what can be delegated.
  • Four gaps became visible: hierarchy of priorities, limits of delegation, goals, and open decisions.
  • The answers go straight into the second brain before the next question follows.
  • The second question after that: what task can you take on for me right now.

Interview prompt card

The prompt that makes the AI ask first and work second, plus five questions and a sequence. PDF · A4, 1 pages, 37 KB, as of 3 September 2026.

This note keeps growing

2026-09-03: Deepened: the four-step sequence, the prompt in full, the four gaps it exposed, and where the technique comes from.

2026-09-02: Planted from the "Agenten-Mindset" Deep Dive, 3 August 2026.

ZukunftBilden GmbH · Salzburg · +43 681 81655313 · office@zukunftbilden.eu