Let’s talk about why “I’m sure” stopped meaning anything.
There was a drawing office I sat near for a while. Old engineer, sixty-something, a tube of A1 prints under his arm, and a rubber stamp with his name and his registration number cut into it. He’d read the sheet. Took his time about it. Then he’d either stamp it or hand it back with two words and a look.
The stamp was the whole business. Not the drawing. Anybody could draw. The stamp meant one specific man had looked at that structure, decided it would hold, and put his name on the decision. If it didn’t hold, his registration was on the line, his insurance was on the line, and he’d be sitting in a room with people who had no interest in his reasoning. He’d have to explain himself to a room that had already decided he was the reason.
That is what certainty used to cost.
Now picture where we are. You open a document and the writing tool offers to finish your sentence. You ask your bank’s chat window whether you can afford the bond. Your kid asks something and gets an answer in half a second, in the same warm, unhesitating voice the machine uses for everything else. Nobody in that list is taking a risk. Not one of them can lose a licence, a fee, a client, a night’s sleep.
And yet they all sound like the engineer with the stamp.
That’s the thing nobody’s pricing. Certainty was never information on its own. It was information because it was expensive to produce. A confident man used to be a man who had something riding on being right. His confidence was a receipt. You could read it, weigh it, and decide whether the thing he was putting up was worth as much as the thing he was promising.
We didn’t just automate the answers. We automated the tone that used to be earned by carrying the consequence.
Think about what a hedged sentence does in a real business. A builder tells you the slab will be fine but the drainage is a problem, and he wants it in writing that he flagged it. He’s annoying. He’s slow. He costs more. But that sentence is worth money, because every qualifier in it is a place where he stuck his neck out. He’s telling you exactly where he can be hurt. That’s not weakness. That’s a map of his honesty.
Now look at what the machine does with the same question. It gives you a paragraph with no neck in it. Fluent, symmetrical, complete. Confidence with nothing behind it, delivered at the speed of typing.
Here’s the part that bites. We’ve started using that fluency as a hiring signal. Procurement teams read polish as competence because for thirty years polish was expensive, and expensive polish meant somebody senior had signed off on the work. We kept the read and threw away the reason it worked. So the vendor with the beautiful deck and the confident guarantee wins the contract from the contractor who says, plainly, “this will hold to about forty thousand units and then the database falls over, here’s the date we think that happens, and here’s what we’d do about it.”
One of those two people can be fired for being wrong. Guess which one gets the deal.
What people are missing is that the problem isn’t that AI is overconfident. Overconfidence in a person is a character flaw. Overconfidence in a system is a pricing failure. It floods a market that was clearing on risk with a supply of assurance that costs the supplier nothing, and when supply is infinite and free, the signal stops being a signal.
You can watch the collapse happen in two directions at once. Some people now believe every confident paragraph they read, because they never learned to ask what it cost the speaker. Others have gone the other way and believe nothing at all, which sounds tough and is actually just as cheap, because refusing to trust anyone costs you nothing either. Both of them have given up on the same skill: reading a stake.
And that skill is the one that keeps you sovereign in a room full of machines. Not detecting AI. Reading who can lose.
I’ll go further. Doubt is the last thing in this economy you can’t fake at scale. Anyone can generate a claim. Almost nobody can generate a number they’d bet the company on, with a date on it, a failure mode named, and a penalty attached. That’s not a rhetorical trick. That’s a different class of statement entirely. It has a floor under it.
So the operator lesson is short and it’s uncomfortable: if you’re selling anything in the next three years, price your own doubt out loud. Say what you’re sure of and what it costs you to be wrong. Put a number on it. Name the thing that would break it. Offer a guarantee with actual teeth, and then be the person who eats the loss when the guarantee gets called. In a market drowning in free certainty, that is the only claim anyone can still hear.
And if you’re buying: ask the question that separates the two. What would make you wrong? If the answer arrives fast, smooth, and free of consequence, you just learned the price of the opinion, which is zero. The good ones pause. The good ones get uncomfortable and start talking about the conditions under which their own answer falls apart. That discomfort is the sound of somebody who has something to lose.
One more thing, and this one is closer to home. The voice in your own head has gone smooth too. You can now get a confident answer to any question about your business, your marriage, your health, before you’ve had the argument with yourself that used to produce your own view. And the machine’s answer will feel certain in a way yours never does, because it has no memory of being wrong in front of anyone.
You have been wrong in front of people. That’s why your certainty is worth something when you finally offer it.
The engineer’s stamp is worth something because the man who owns it can be sued. The name on the drawing is worth something because the name has a house and a family and a reputation attached to it, and a bad year if the thing comes down.
Keep that in the room. Whatever you build, whatever you sell, whoever you let make the call. Not the most fluent voice. The one with the most to lose.
That’s the whole test, and it’s older than any of this. When someone tells you they’re sure, ask what it costs them to be wrong. If the answer is nothing, you’re not hearing certainty. You’re hearing a machine, or a man who has learned to sound like one.
You’ve got this.