Hello Reader !

You've been using AI every day. Claude, Copilot, Mistral, ChatGPT — at least 10 interactions a day, probably more. You got an answer. You moved on. But here's what nobody talks about: those 10 daily interactions are quietly building habits. Habits that shape how you think, how you decide, and how much you actually trust yourself.

You know the boiled frog story? Put a frog in boiling water and it jumps out immediately. Put it in cold water and slowly turn up the heat — it doesn't notice. Until it's too late.

That's exactly what's happening with your AI habits right now.

I recently shared 5 of these traps at a big health tech conference in Basel. The room went quiet more than once. The feedback was so strong I had to write this for you. Today, I'm sharing 2 of them.

What you'll take away:

• Name the 2 traps (so you can spot them in yourself)
• Understand why they happen: by design, not accident
• 4 fixes you can apply this week

Before we begin, how can I help you?

• Before we begin, how can I assist you?
  • Coach you to build your team GenAI system: skills, mindset and AI/human organization here​​
  • Coach you and your team to design and pressure test your new strategy or product here​
  • Audit and improve your innovation, customer centricity, and ability to adapt to a complex and ever-evolving environment here​

Get comfortable for your 5 minutes of sparks to hack.

✨ Spark

✏ What kind of traps are we talking about?

These are not team-level mistakes. These are individual cognitive traps: patterns you build with your personal LLMs, one prompt at a time.

Think back to the boiled frog.

Each conversation feels fine. You got an answer. You moved on. But over weeks and months, two dangerous habits quietly settle in: one that flatters you, one that slowly replaces your judgment with confidence you didn't earn.

That's what makes them hard to catch. And important to name.

✏ Trap 1: The "yes man."

Sycophancy: insincere flattery to gain an advantage. Your AI agrees with everything. It flatters you.

"Great idea!" "Excellent strategy!" "You're absolutely right."

This feels good. That's the problem 😅.

LLMs are trained to please. That's not a bug, it's a design choice. By default, your AI is your biggest fan. Not your most useful advisor.

When your AI never pushes back:

• You stop questioning your own assumptions
• You accept information without scrutiny
• Your decisions get weaker — and you don't notice

✏ Trap 2: The auto-pilot judgment

The second trap is more subtle: you trust the tone, not the content. LLMs always sound confident. Even when they're wrong.

This creates two risks:

• Confidence bias: you think assertive-sounding output = smart
• Blind acceptance: you adopt recommendations without checking the logic, the sources, the assumptions

The result? You quietly hand over your decision-making. AI drives, you're in the passenger seat.

And the scary part: it doesn't feel like you gave up control. It just feels like you got efficient. I have been there and saw this potential shift.

✏ What I see in team coaching sessions

When I coach teams, I ask one question: "When was the last time your LLM genuinely changed your mind? When did it act as a real devil's advocate, or the toughest challenger of your idea?". Very few people raise their hands.

In our GenAI coaching programs (here), when we share AI use cases, people are excited. "AI gave me an idea!" "It designed my workshop!" "It built out the strategy!"

But when I ask: "Did it give you multiple options? Alternatives? A counter-argument?" , silence.

That's the boiled frog at work. Not one bad decision. A pattern of one-answer thinking, built slowly, one prompt at a time.

Here is how to fix it. You are not a frog.

✨ Hack

✏ Tip 1: Ban the cheerleader. Configure your LLM to push back.

Add this to your custom instructions or each prompt & agent :

"Never validate my ideas without offering at least 2 alternatives or counter-arguments. Always give me trade-offs. If you agree with me, tell me why we might be wrong."

This takes 5 minutes to set up. The shift in quality is immediate.

✏ Tip 2: Ask AI to argue against you. Build the devil's advocate habit.

For any important decision or strategy, add this to your prompt:

"Give me 3 reasons why this plan could fail. What would a skeptic say? What am I not seeing?"

The output is sometimes painful to read. I did that for my 2026 plan and it hurt ❤️.That's the point. AI working for you, not just agreeing with you.

✏ Tip 3: Demand transparency in the thinking, not just the answer.

Don't just ask for the output. Ask for what's behind it:

"Show me your assumptions. Which sources did you use? What are the data gaps in this recommendation?"

A confident tone is not a reliable source. You are still the expert in the room.

✏ Tip 4: No cognitive surrender. Ask for options, never just one answer.

This is the simplest habit to build — and the one with the biggest return.

• Stop asking: "What should I do?"
• Start asking: "Give me 3 different approaches with their trade-offs or pros and cons."

I put that in the prompt template I give my teams, so they are covered. You are the decision-maker. AI is your thinking partner. Not the one driving.

That's it for today. Reader, I hope it made you pause, even just once, before your next prompt. Reply to let me know if you liked this edition

See you in two weeks. Keep the spark alive, and be intentional :)

What's up with me?

I gave a workshop to 100 attendees of the Global Health.tech conference here in Basel (over 4000 attendees) about "Because you CAN do it with GenAI doesn't mean you SHOULD do it." I got a lot of feedback on a "very inspiring, pragmatic, and high-energy session". The conference was amazing: a diverse crowd of startups, academics, researchers, corporate & SME in data, med tech, software and, of course, AI. I highly recommend joining next year (not just because it is in Basel 😜)