The Reverse Centaur Problem

You pull up a deal in your CRM and the AI score says 82% likely to close. You didn't ask how it got there. You didn't need to — it sounded certain, and certain is contagious.

That's not a story about a bad rep. It's the default now. AI hands you a score, a forecast number, a next-best-action, a drafted email — and your job quietly shifts from deciding to approving. The work still looks like judgment. It isn't anymore.

What's Actually Happening

This has a name: EY calls it the "reverse centaur." In a 2026 piece on AI and decision trust, EY Insights' Patricia Camden traced it back to competitive chess, where the original "centaur" model — human and machine collaborating — consistently beat either one alone. AI did the heavy lifting; people decided what mattered.

Enterprise AI is starting to flip that. As Camden puts it, AI increasingly "sets the pace while people review, approve and occasionally challenge its recommendations." Her research is blunt about why teams struggle with this: it's rarely that the model is wrong. "The issue isn't intelligence," she writes, "it's confidence."

Why This Keeps Happening

Here's the part I keep coming back to: confidence and accuracy are not the same signal, and a fluent AI output makes them feel identical. The fix isn't distrust — checking everything just turns AI into more work, and the productivity gain disappears either way. It's being deliberate about which claims are safe to move on and which ones need a second look before they touch your forecast.

What This Costs You

If you're an AE, this shows up as a deal that was "definitely closing this month" according to the score, until it wasn't — and now you're explaining commit slippage in a forecast call you were confident about a week ago. The score didn't lie. You just never asked what it was actually based on.

If you manage a team, it compounds. Five reps each trusting a score they didn't interrogate rolls up into a number you present with real conviction — right before it falls apart in the room. By the time it shows up in a QBR, it's not a coaching conversation anymore. It's a credibility problem.

What Good Judgment Actually Looks Like

I've watched this play out both ways, long before AI was scoring anything.

A rep I worked with called a deal "warm" for three straight forecast calls because the buyer nodded through every demo. Nodding isn't a signal. It's politeness. Nobody had asked the buyer what would need to be true for a signed contract by end of quarter — so nobody knew the real answer was "budget isn't approved yet," until it cost the deal.

A manager I sat with ran the same pipeline review every Monday, asking the same three questions, and closed every call having updated exactly nothing about what he actually believed. The ritual felt like diligence. It wasn't — it was a habit standing in for one.

Now add an AI-scored deal to either scenario and the pattern doesn't change, it just moves faster. A tool says 82% likely to close. That number came from something — deal velocity, engagement signals, similar-deal history — and none of it tells you whether the buyer's actual blocker got solved. The rep who pauses to ask "what is this score actually based on, and is that still true" catches the gap before it hits the forecast. The rep who doesn't finds out in the deal review, in front of their manager, why the number moved.

Try This In Your Next Pipeline Review

Next time you present a deal's score, say the evidence out loud before you say the number.

"This is scored at 82% because Sandra confirmed budget on the 12th, and the contract's already out for signature."

If you can finish that sentence with a real name, a real action, and a real date, lead with the score — you've earned it. If you can't, say what you actually know instead: "The AI has this at 82%, but the only thing I can confirm is the demo went well."

Either way, you've just turned a number nobody can question into a claim your manager can actually coach you on.