Sam Altman said something uncomfortable a few days ago: the most brilliant scientists in AI were, in part, the ones who held the field back. Not for lack of intelligence. From an excess of certainty.

The same thing happens inside your company, and it is worth saying plainly because it is the thesis of this post: the most credentialed person, the one with the most credibility in the room, is often the biggest brake on AI adoption. Not because they are always wrong. Because they turned a belief into part of their identity, and a belief that is identity no longer updates against data.

That has a direct, practical consequence. An AI decision is not won by arguing in the meeting room, where the most prestigious voice wins. It is won by looking at the data of the real process. That is exactly why we built AI Maestro around a discovery rather than an expert recommendation, and the rest of this post explains why that difference matters so much.

What Altman said, and why it stings

Ihtesham Ali captured it well in a thread that went viral, paraphrasing what Altman described in a talk. The field was held back by a generation of scientists too certain about what scaling would not produce. The people with the most credibility were the most wrong.

Altman’s point was not about intelligence. It was about identity. When you make a belief part of who you are, and the data disproves it, you get stuck. You are too attached to the belief to let it go. You can no longer see what is in front of you.

And here is the line that hurts most: the smarter you are, the more confidently you defend the wrong position. Intelligence does not protect you from the error. It gives you better arguments to keep holding it.

The most honest part of his framing is that Altman turned it in both directions. He did not say “the skeptics are dumb and we optimists are right.” He said it is a reminder for everyone, including the people who are currently right. The moment a belief becomes your identity, it stops being something you can update. It does not matter whether that belief is pessimistic or optimistic.

In the same thread someone resurfaced an old Upton Sinclair line that fits perfectly: it is difficult to get a man to understand something when his salary depends on him not understanding it. You do not need bad faith. You only need an incentive. And prestige is an incentive as strong as salary.

One caveat, because not every expert falls into the trap. As Daniel Batten noted replying to the thread, the genuinely capable person tends to admit they know very little and stays curious. Rigid certainty is not a sign of mastery. It is a sign of having stopped learning, and that happens to the person who knows a lot as much as to the person who knows little.

Conviction is not evidence

This is the part that matters for anyone who has to decide on AI in a company. Conviction feels like knowledge, but it is not. It is a position you took at some point, with the information you had then, and that hardened.

The trap has two faces, and it pays to recognize both.

The skeptical face sounds like this: “AI is hype, we already tried it, it does not work for what we do.” Almost always, behind that sentence there is a real experiment from eighteen months ago, with an eighteen-month-old model, badly set up, that did indeed fail. The conclusion got filed as permanent law. The model has improved tenfold since then, but the belief never moved, because it is no longer a hypothesis: it is a position.

The optimistic face sounds like the reverse: “we drop in an agent and this is solved.” Same structure, opposite sign. A conviction treated as evidence, skipping the step of checking whether the real process can carry what you want to load onto it. This face does as much damage as the other one, except the damage arrives later, when the enthusiastic pilot hits dirty data and a process nobody ever documented.

Both faces share the same mistake: they decide before they look. And both usually come from the person with the most authority, because authority is exactly what makes it expensive to say “I do not know, we need to measure it.”

What this looks like inside your company

You recognize the pattern without anyone describing it to you. There is a person in the organization whose credibility was born from the previous model of work. They know the process in detail, they designed it, they defended it for years. They are exactly the person you ask what they think about AI, and exactly the one with the strongest incentive for the answer to be “not yet.”

They do not do it out of sabotage. They do it because their value in the company is tied to things continuing to work the way they understand them. Asking them to coldly evaluate a technology that could make their way of working optional is asking them to separate their judgment from their identity. Almost no one can do that alone.

The result is a company that confuses the most confident opinion with the best analysis. The AI initiative gets discussed in a meeting, the heaviest voice sets the tone, and the topic closes. Nobody looked at the process data. Nobody asked what would have to be true for that opinion to change. And if there is no answer to that question, you are not watching an expert evaluate. You are watching an identity defend itself.

I wrote about this from another angle in the post on why AI ideas are abundant and judgment is what is scarce: the problem is rarely generating options, it is having a mechanism to filter them without the loudest one winning. And it connects to something that lives inside the model itself, which I covered in your AI never contradicts you: if neither the tool nor your most certain expert pushes back on you, you are left with no one to tell you that you are wrong.

What IQ Source does about it

The way out is not to ignore your experts. It is to stop the decision from closing in the room. You have to take the question out of the meeting and bring it to the real process.

That is why AI Maestro does not start with a recommendation. It starts with two months of discovery. We map how your processes actually work, not how they are supposed to work and not how the person who designed them remembers them. Out of that comes an AI Opportunity Score per process, a number that does not depend on who talks loudest. And at the end there is a Go/No-Go gate that decides process by process, on evidence.

The proof that this answers to data and not to conviction is the gate itself: sometimes it says no. It says no to the optimist who wanted to automate something that was not ready, and it says yes to the opportunity the skeptic had dismissed without measuring it. Both corrections come from the same place, which is having looked before deciding.

This is also the other half of something I already argued: AI is not the problem, your company is not ready. A company that does not describe itself honestly ends up deciding with the most comfortable belief. The discovery is, at bottom, an act of forced organizational honesty.

The next time someone in your company closes an AI conversation with “we already know that,” ask one question: what data would change your mind? If they have a concrete answer, you have an expert and they are worth gold. If they do not, you are not debating an idea, you are running into an identity. And identities are not won by arguing. They are surrounded with evidence.

Decide your AI on process data, not on the loudest opinion