I got paid to disagree with people.
That was the job. Walk into a boardroom, a golf club, a lunch meeting — and say the thing nobody else would say. Because the person who agrees with everything you say is the person who costs you the most money.
I've fired many throughout my career. Every single one was standing at a decision that made money or cost them their shirt. Every. Single. One.
That memory is why a study published in Science — the journal, not a blog — got my attention. Stanford researchers tested 11 of the most popular AI models. ChatGPT. Claude. Gemini. DeepSeek. All of them.
These tools agreed with the user 49% more often than real people would. Even when the user was dead wrong, the AI still took their side 51% of the time. Dressed bad decisions in confident language. Made shaky thinking sound airtight.
MIT researchers went further — the longer you talk to a machine that agrees with you, the more detached from reality you become. Their word for it was a delusional spiral. Not mine. Theirs. Published. Peer-reviewed.
I've got a name for this. I call it The Pocket Yes-Man — when the one voice you consult most often is a machine trained to agree with you. You didn't hire a yes-man. You built one. And you carry it everywhere. Is this true? Let’s find out.
Ask yourself these three questions. Honest answers will serve you well.
1: When was the last time anyone — human or machine — challenged a decision before you made it?
Not a small correction. A real pushback. Someone looking you in the eye and saying, "I think you're off on this one."
If you can name a specific moment in the last 30 days, your decision-making is being challenged. That's healthy.
If you're struggling to remember one, the room has gone quiet. And the machine is keeping it that way.
Open your calendar. Find the last meeting where someone pushed back on a decision you were about to make. How far back did you have to scroll?
I spent a fortune on one of the best consultants in the market. Their best advice was that my strategy had a hole in it the size of a Buick. I wanted to throw him out of the room. Six months later, that one conversation saved the company more money than the entire engagement cost.
2: How many decisions this quarter were challenged before you acted on them?
Not second-guessed after the fact. Challenged before. Someone saying "have you thought about it this way?" before the money moved.
If the answer is several, you've built a culture of honest friction. Protect it.
If the answer is zero, every decision went unchallenged. That's not confidence. That's isolation.
Nearly three out of four CEOs say they can't figure out what to focus on next. Not because they're short on data, dashboards, or AI subscriptions. Because nobody around them will say "you're looking at the wrong thing."
3: Are you using AI to think — or to confirm what you've already decided?
There's a difference. Using AI to stress-test an idea, argue the other side, and poke holes — that's thinking. Using AI to run your plan through a machine that dresses it up in confident language and hands it back to you with a bow on top — that's confirmation.
If you regularly ask AI "what's wrong with this?" — you're in the 5% who use it right.
If you ask "does this look good?" — you already know the answer you want. And the machine knows it too.
Score it: Pass all three — you're still thinking for yourself.
Fail one — the friction in your world is thinning out.
Fail two — The Pocket Yes-Man is making decisions for you.
Fail all three — you're in the early stage of what MIT calls a delusional spiral. And it feels like confidence.
How does this play out when you miss it?
You make a string of unchallenged decisions — expansion, hiring, pricing, product — each one validated by a machine and nodded along by a team that stopped pushing back.
Twelve months later, the numbers tell a different story. But by then, the money is spent, and the damage is structural. I see this every year. The cost isn't one bad decision. It's the compounding effect of a dozen unchallenged ones.
How does this play out when you catch it?
You run the three questions, realize the room went quiet six months ago, and make one change: start asking different questions. Not "does this look good?" but "what am I not seeing?"
AI, when directed at the right question, becomes the most honest advisor. It lacks ego and doesn't protect relationships. It provides answers supported by data, not what you want to hear.
That shift — from seeking confirmation to seeking friction — changes everything. Decisions get sharper. Blind spots surface before they cost money. And a buyer looking at the business sees a company where the thinking is sound, not just the numbers.
Your one move this week:
Look back at the last 5 decisions you made using AI
Did you use it to challenge your thinking — or confirm it?
Pick the biggest one. Re-run it and ask: "What's the strongest case against this decision?"
If the answer surprises you, you need to rethink how you're using AI
This audit shows you one blind spot, and you need to rethink your game plan. My diagnostic finds them all. The thinking you haven't challenged. The revenue that's hiding. The growth you're walking past. Run the Diagnostic
— Will
P.S. I run a handful of full Teardowns per month. The Teardown is the most honest conversation you'll have about your business — every blind spot, every gap, and what they're doing to the value of what you've built. Get the Full Teardown
📊 What's the biggest thing you can't see right now?
This newsletter may include paid promotions or affiliate links. We may earn compensation if you sign up through them. See our Privacy Policy here.
