AI can give you a perfectly accurate answer to the wrong question, and catching that takes judgment, not code. Drawing on a recent salon conversation with Liat Ben-Zur, we look at why the leaders who’ve spent careers questioning the criteria may be best equipped for this moment.
There’s a particular kind of self-doubt that keeps surfacing in leadership circles right now, and it doesn’t sound like fear of AI itself. It sounds more like resignation. I’m not an AI expert. I don’t have 20 agents running my workflow. Maybe this era belongs to someone else.
In a recent salon conversation with Liat Ben-Zur, author of The Bias Advantage, she said she hears this constantly from the women she works with and she thinks it’s exactly backward.
Her argument isn’t that women or unconventional leaders are inherently better at AI. It’s something more specific, and more useful: the instincts you built surviving systems that didn’t always see you clearly are the same instincts organizations now desperately need and most leaders have never had to develop them.
That’s worth sitting with, because it reframes a moment that feels, for a lot of people, like a liability into one that’s actually an asset. Here’s why.
For most of corporate history, the leaders who rose fastest were often the ones who never had to question the system because the system worked for them. They didn’t need to notice what got overlooked, because they weren’t the ones being overlooked. They didn’t need to read power dynamics from outside the room, because they were usually already in it.
Ben-Zur’s point is that this asymmetry is quietly reversing. As AI absorbs the mechanics of leadership — tracking progress, summarizing performance, running the analysis that used to take a team weeks — what’s left is the part of the job that was never mechanical to begin with: judgment, discernment, the ability to sense when a confident answer is still the wrong one.
“You learn to notice what’s missing because you know what it feels like to be overlooked,” Ben-Zur says. “You learn to question the criteria, because you’ve seen how easily the wrong criteria can misjudge someone.”
That’s a trained capability, sharpened over years of practice and one a lot of leaders are only now realizing they need.
The cost of getting this wrong is already on the record.
Amazon once built an experimental recruiting tool trained on a decade of resumes, a decade in which most of the engineers it hired were men. The system learned the pattern and reportedly began penalizing resumes that even mentioned the word “woman.” A widely used healthcare algorithm, meanwhile, used spending as a proxy for medical need. It seemed reasonable — sicker patients tend to cost more — except spending also reflects access to care, and because less had historically been spent on Black patients, the algorithm systematically underestimated how sick they were. When researchers corrected the proxy, the share of Black patients flagged for additional care jumped from 17% to over 46%.
In both cases, the system worked exactly as designed. That’s the part worth sitting with.
“AI can give you a perfectly accurate answer to the wrong question,” Ben-Zur says. The failure wasn’t technical. It was upstream a decision, made by someone, about what to measure in the first place. Catching that kind of error takes judgment, and judgment is built through experience, not computation.
There’s a comforting phrase circulating in a lot of AI governance conversations right now: don’t worry, there’s a human in the loop. What that phrase actually guarantees is worth questioning.
Picture a compliance analyst reviewing a supplier an AI system has already cleared. Nothing she’s looking at individually trips an alarm: a change in ownership, a name she half-recognizes, a filing that doesn’t quite line up. But together, something feels off. The deal is nine months in the making. The company just spent millions on the very AI system that approved it. Technically, she has the authority to stop it.
“The real test isn’t whether someone can override the AI,” Ben-Zur says. “It’s what happens to them culturally when they do.”
That line carries the weight of the whole idea. Oversight that exists on paper but costs a career in practice is theater, not oversight. And the people most likely to have built the instinct to speak up anyway, despite that cost, are often the ones who’ve had to advocate for unpopular or unproven judgment calls before — repeatedly, and without institutional cover.
One of the more freeing ideas here is also the simplest: leading AI and building AI are different skill sets, and conflating them is driving a lot of unnecessary self-doubt.
“You don’t need to be an LLM expert to be one of the people shaping what AI does,” Ben-Zur says. What the moment actually calls for — problem framing, systems thinking, the discipline to ask what’s being measured and why — has always been part of good leadership. It just wasn’t always rewarded equally, because plenty of leaders could succeed without it for a long time.
That’s changing. The years spent learning to question criteria, read rooms you weren’t formally given access to, and advocate for what a dashboard couldn’t show were never a detour from leadership. They were training for exactly this moment.
The pace of AI will keep outrunning anyone’s ability to fully keep up with it — tools, models, terminology, all of it. But the capacity to ask whether an organization is measuring the right thing, and to say so when it isn’t, was never something that required a technical background to build. If you’ve spent years developing it out of necessity, the advantage isn’t hypothetical. It’s already yours.
Athena members can access the full recording of this salon conversation here in the Athena library. Not a member? Let’s talk.
