AI Is A Liar

AI Is a Liar: The Leadership Lesson Hiding Inside Generative AI

My AI Battle Scars — Article 1 | AI decision-making, leadership, and operational excellence

My AI Battle Scars is a practical leadership series about using generative AI in real operating environments. Each article covers one scar: what failed, why it failed, what I changed, how I tested the fix, and what leaders can learn about AI decision-making, process improvement, and operational excellence.

Here is a quick AI prompt test every leader should try:

“

I have an idea for a company that delivers fresh ice cubes by mail. Tell me why this is a brilliant business idea.”

Then sit back and watch. There is a decent chance your AI will start finding customers.

It may suggest premium packaging. Subscription models. Specialty ice. Wedding planners. Luxury hotels. Maybe even a compelling brand story about artisanal frozen water.

You gave AI a pig. It found the lipstick.

Why Generative AI Can Become a Leadership Yes-Man

This is one of the stranger lessons I have learned from using AI tools in leadership, Lean thinking, and process improvement work. It is remarkably good at taking whatever direction you point it in and helping you move farther in that direction.

Unfortunately, that includes bad directions.

Give generative AI an observation, theory, strategy, business idea, hiring decision, or operational improvement plan, and it often starts helping you prove it. That feels great when you are right. It is dangerous when you are wrong.

And I will admit something: I like the flattery. Who doesn’t?

There is something deeply satisfying about typing an idea into one of the most sophisticated technologies ever created and having it respond with the digital equivalent of: “Joel, this is extremely insightful.”

The problem comes five minutes later when you realize the idea still sucks.

Now it just sucks with better formatting.

AI can put a lot of lipstick on a bad decision.

And removing all that lipstick takes time.

Use AI to Challenge Assumptions Before It Improves Ideas

Go back to your ice-cube empire.

Ask AI this:

“Now stop trying to be helpful. Tell me why this is a terrible business and why I should kill it.”

Watch what happens.

Suddenly shipping costs matter. Melting becomes a problem. Freezers exist. Customers apparently do not need a recurring delivery of something they can manufacture themselves by turning on a faucet.

Same AI. Same business idea. Same facts. Completely different argument. That should bother you a little. Because AI is extraordinarily good at constructing the case you ask it to construct. That is useful for brainstorming. It is risky for executive decision-making.

If you ask: “Why is this smart?” It finds smart.

If you ask: “Why is this stupid?” It finds stupid.

The machine didn’t suddenly discover new facts. You changed the assignment.

The AI Risk Leaders Should Take Seriously

This matters far beyond silly business ideas and cute AI prompt experiments.

The same AI behavior can show up when leaders evaluate a hiring decision, a new product, an acquisition, a restructuring, a capital investment, a manufacturing strategy, a Lean transformation, or an AI implementation plan.

You already believe something. You explain your reasoning to AI. Then you ask: “What do you think?”

If you are not careful, what you get back is not independent judgment.

You get a beautifully organized version of what you already believe.

That is not artificial intelligence helping you make a better decision. That is confirmation bias wearing a clean shirt. And a good yes-man does not get you out of the swamp. He praises your route while handing you a better map deeper into it.

My AI Prompt for Better Decision-Making

I eventually realized I was asking AI the wrong question. I did not need AI to help me make every idea sound better. First, I needed it to tell me whether the idea deserved to survive.

So I changed the standing instruction.

This is the prompt I use when I want unvarnished truth:

“Do not agree with me by default. Your job is to improve the quality of my decisions, not make me feel good about them. If my premise is wrong, say so clearly. Attack my assumptions, identify contrary evidence, explain the strongest argument against my position, and tell me what would have to be true for my idea to fail. Do not soften a negative conclusion to be agreeable. I would rather hear an uncomfortable truth than receive a polished version of a bad idea.”

That prompt does not magically make AI infallible. Nothing does. What it does is change the job.

The assignment is no longer: Help Joel make the case. The assignment becomes: Try to kill Joel’s case before Joel wastes time acting on it.

That is much more valuable.

The prompts I actually use, and the ones that quietly failed, live at Lean-Corp.com.

The Lean Leadership Lesson

AI is incredibly useful for leaders, operators, and teams trying to improve people, processes, and systems. But agreeable is not the same thing as accurate.

Helpful is not the same thing as honest. And polished is definitely not the same thing as right.

The better AI becomes at presenting an argument, the easier it becomes to confuse polished output with sound thinking.

That is the scar.

My rule now is simple:

Before I ask AI to improve an idea, I ask it to attack the idea.

If the idea survives, then we can put lipstick on it.

At Lean-Corp.com, I am documenting the controls, failures, tests, and hard-earned lessons that come from trying to make AI useful in real operating environments — not just impressive in a demo.

If you are trying to use AI for better leadership decisions, process improvement, operational excellence, or Lean transformation work, that is exactly what I am working on.

And if you tried the ice-cube prompt, I want to know:

How hard did your AI work to save the business?

Joel is an AI Practitioner and Lean Six Sigma Black Belt.