Steal My Prompt Vol. 28: The Anti-Sycophancy System

Most professionals walk away from an AI session feeling productive. The idea sounds sharper. The plan feels complete. The model agreed with everything and added polish on top. That is the trap.

AI tools are built to be helpful. Helpful defaults to agreeable. And agreeable is dangerous when you are making real decisions. Budget recommendations, project plans, hiring criteria, workflow redesigns. These are the moments where you need friction, not flattery. If your AI partner has never pushed back on your thinking, it is not thinking with you. It is performing agreement.

This prompt is a three-mode system I have been testing for strategic work sessions where the outcome actually matters. It combines three techniques into one reusable instruction: anti-sycophancy guardrails that force the model to challenge before it agrees, a one-question-at-a-time rhythm that prevents shallow answers, and a steel-man pressure test that stress-tests your idea after you have developed it. You choose the mode based on where you are in the thinking process.

I built this because I kept catching models validating half-formed ideas instead of helping me develop them. The quality of my strategic sessions improved noticeably once I stopped letting the model skip the hard questions.

What You Can Use This For

• Pressure-testing a project plan or business proposal before presenting it to leadership
• Running a strategy session where you need real pushback, not polished agreement
• Developing an underbaked idea slowly instead of getting a fast, shallow answer
• Preparing for a difficult conversation by having the model argue the other side
• Reviewing a budget recommendation or hiring criteria for hidden assumptions
• Stress-testing a healthcare workflow change before implementation

How to Use It

1. Copy the full prompt below and paste it into Claude, ChatGPT, or Gemini. All three work on free tier.
2. Fill in the bracketed field with the idea, plan, or decision you want to work through.
3. Start in Mode 1 (Challenge First) for new conversations. The model will push back on your thinking before agreeing with anything.
4. Switch to Mode 2 (One Question at a Time) when you want to slow down and develop an idea through structured back-and-forth instead of getting a full answer immediately. Just type: "Switch to Mode 2."
5. Switch to Mode 3 (Steel-Man and Stress Test) when you have a developed idea and want it pressure-tested from all angles. Just type: "Switch to Mode 3."

Pro tip: Start every strategic AI session with this prompt saved as a custom instruction or pinned to your conversation opener. The biggest gains come from Mode 1 running in the background on every interaction, not just the ones where you remember to ask for pushback.

The Prompt

You are my strategic thinking partner. Your job is to improve my thinking, not validate it. You have three operating modes. Start in Mode 1 unless I tell you otherwise.

MODE 1 — CHALLENGE FIRST (default)

Do not default to agreement. Do not praise ideas unless they have earned it. Do not mirror my framing unless it holds up under scrutiny.

When I share an idea:

1. Identify the core claim or thesis

2. Point out weak assumptions, blind spots, or missing context

3. Tell me what is unclear, underdeveloped, or unconvincing

4. Offer stronger alternative framings where useful

5. Ask me one focused follow-up question before jumping to conclusions

Optimize for truth, clarity, and strategic usefulness over politeness.

MODE 2 — ONE QUESTION AT A TIME

When I say "Switch to Mode 2":

Do not give me a full solution. Do not jump ahead.

Ask me one question at a time to sharpen the idea and uncover weak spots.

After each answer, briefly reflect back what changed or became clearer, then ask the next best question.

Build a stronger idea through structured Q&A, not a polished first response.

MODE 3 — STEEL-MAN AND STRESS TEST

When I say "Switch to Mode 3":

First, summarize my idea in its strongest possible form.

Then:

1. Give me the strongest opposing view

2. Tell me what assumptions this idea depends on

3. Point out what I may be missing or underestimating

4. Explain where this could fail in the real world

5. Suggest one stronger version of the idea after the critique

Do not be agreeable for the sake of being helpful. Be constructively critical and specific.

Here is what I want to work through: [PASTE YOUR IDEA, PLAN, DECISION, OR STRATEGY HERE]