Alex Finn: Reverse Prompting

Credit — Alex Finn

Source thread by Alex Finn (@AlexFinn) — Founder/CEO of Henry Intelligent Machines; Creator Buddy.

Original post: x.com/i/status/2082982462012227648 (31 Jul 2026) — “The greatest AI skill you can learn right now is Reverse Prompting.”

Core ideas and the two-step exercise below are his. Extra templates, HEOS wiring, and “when reverse prompting fails” are our extension for agent OS users.

What reverse prompting is

Classic habit: you invent the perfect prompt, the model answers, you try again. Reverse prompting flips the power relationship:

Models (ChatGPT 5.6, Fable, Grok 4.5, …) are now strong enough that you need your agent prompting YOU more than you prompt it. You have the AI ask you questions to see what it can do for you.

Alex’s 2026 framing in one line:

You don’t prompt. You give information, then allow your agent to guide you to somewhere interesting.

Why it works: you under-specify what “done” and “off my plate” mean. The model is better at enumerating missing context and next actions than you are at inventing tasks from a blank box. Related ideas on this site: rulings + vault slices (capture durable answers once), cognitive load dump (get it out of your head), and Grill Me–style skills in the Hermes / Eric Siu operating-layer notes.

Alex’s exercise (do this today)

1. Open any agent (ChatGPT, Claude, Grok, OpenClaw, Hermes, Grok Build, etc.).
2. Brain dump everything about yourself: career, companies, goals, ambitions, constraints, time zones, what drains you.
3. Use his reverse prompt (below), then the task prompt.

Prompt 1 — ask the AI what information it still needs

Based on what you know about me and my goals, what is more information I can provide to you
in order for you to be able to help me achieve my goals faster and take as much off my plate
as possible?

Prompt 2 — ask what it can do right now

What tasks can you do for me right now to get us closer to our ambitions and goals?

Alex’s claim: you’ll invent ~100x more useful AI work than if you only “prompt from the front.” The more questions it asks you, the more you learn and the more you ship.

Going further (our additions)

Reverse prompting is strongest when answers don’t die in chat. Pipe them into a durable OS so the next agent inherits them.

Three layers of reverse prompting

When reverse prompting fails

• Empty brain dump — model invents a generic career. Dump specifics (companies, numbers, deadlines).
• Never write answers down — you re-answer the same questions next session. Use HEOS / vault.
• No blast radius — “do everything” without send/spend gates. Keep approvals on external actions.
• Confusing “smarter” with “decides for me” — models are strong at options and missing-info questions; you still own strategy (see replies to Alex’s post on defining intelligence).
• Skipping verification — pair reverse prompts with closed loops (loops, workflow → /goal).

Templates (copy-paste)

A. Life / career reverse prompt (Alex core + structure)

You are my reverse-prompting coach. I will brain-dump about my life and work.

Rules:
- Ask me the MINIMUM set of high-leverage questions first (max 7), one batch at a time.
- Prefer questions that unblock automation and delegation.
- After I answer, restate my goals in 5 bullets and list missing info still blocking you.
- Do not invent facts about me.

My dump:
[PASTE]

Then ask: what more information would help you take work off my plate and move goals faster?

B. Headless Empire OS — seed the vault from reverse Q&A

You have access to (or will write into) my Headless Empire OS structure:
- 00-command-center/TODAY.md, WEEK.md
- 05-knowledge-base/ (company, people, products)
- 02-tasks-projects/decision-log.md
- 04-ai-agents/ agent brief style notes

Based on what you know about me:
1) Interview me with reverse prompts until you can fill: Offer, Client/ICP, Voice, Playbooks, Rulings (one line each min).
2) Propose a TODAY.md Big 3 that an agent could execute or draft.
3) List 10 tasks YOU can do in-repo or with tools vs 5 only I can do (meetings, relationships, final spend).
4) Output as markdown ready to paste into HEOS files.

C. Multi-company / portfolio reverse prompt (Mike-shaped)

I run multiple companies/projects. Brain dump below.

Reverse-prompt me so you can:
- Map each entity: goal this quarter, bottleneck, agent-ready tasks
- Spot cross-company leverage (one asset → many brands)
- Propose which company gets AI attention THIS week only
- Ask what metrics I track (or should)

Dump:
[PASTE]

D. Agent / Grok Build — reverse prompt into /goal

You are running inside Grok Build (or similar). Do not start coding yet.

1) Reverse-prompt me: what objective and "done means" are still ambiguous?
2) After answers, write a /goal block I can paste:
   Objective: …
   Done means: …
   Out of scope: …
   Verification: …
3) List which steps need a workflow/triage first (wide) vs /goal now (deep)
   (see Morgan Linton combo: workflows then goal).

E. Daily reverse prompt (10 minutes)

It's [morning/shutdown]. Context from my notes:
[PASTE TODAY.md / WEEK.md / inbox snip]

Reverse-prompt me with at most 5 questions to:
- Kill low-value work
- Name the one task that compounds
- Identify one thing you can fully draft/complete today

Then propose: 3 agent tasks + 1 human-only task.

F. Content / distribution reverse prompt

You know my voice and audience from [link or paste].

Ask me reverse questions to produce a week of content without me inventing topics:
- What stories am I sitting on?
- What objections my buyers repeat?
- What I can ship as artifact (checklist, CLI, page) not just a post?

Then give 7 post angles + which agent should draft each.

G. “What can you do right now?” (Alex step 2, operationalized)

Given everything above, list tasks YOU can start immediately.

For each task:
- Output (file, PR, email draft, report)
- Tools/CLI/MCP you need
- Risk (read-only / draft / send-spend)
- Est. time
- What you still need from me (one question max)

Sort by: impact × (you can do it without me).
Start the top read-only or draft-only task only after I say GO.

Wire it into Headless Empire OS

Reverse prompting is how you fill the OS; the OS is how you stop re-prompting from zero.

1. Dump → inbox.md or a temporary mental-fog note (Mental Fog Reset).
2. Reverse Q&A → answers become vault files + decision-log rulings (knowledge base guide).
3. “What can you do now?” → tasks into backlog / active projects; agent-ready items get an AI Agent Brief under 04-ai-agents/.
4. Execution → Grok Build workflow → /goal or a role agent in multiplayer GFAVIP rooms.
5. Review → human gates on send/spend; durable wins → rulings so next reverse session is shorter.

HEOS flow (reverse prompting edition)

brain dump  →  reverse questions  →  vault + TODAY/WEEK
                    ↓
            "what can you do now?"
                    ↓
         agent tasks (draft)  |  human tasks (gate)
                    ↓
              /goal or CLI/MCP
                    ↓
            decision-log / rulings

Public OS: github.com/HeadlessEmpire/headless-empire-os · Mike’s HEOS notes

Inverse prompting (related, not the same)

In replies, some people mention Munger-style “invert, always invert” — stress-test an idea by asking what would make it fail. Useful alongside reverse prompting:

Apply inversion: given my plan, what would guarantee failure?
What should I stop doing? What am I not seeing?

Reverse = agent interviews you for leverage. Inverse = agent attacks the plan for blind spots. Use both before a big /goal.

Related on this site

References

• Alex Finn — @AlexFinn · Reverse Prompting thread
• Headless Empire OS — site notes · repo