AI Won't Save Weak Judgment in Leadership

A study out of Harvard Business School and UC Berkeley put an AI business advisor in the hands of 640 entrepreneurs and tracked what happened to their revenue and profit. The expectation was a rising tide. What they found instead: the strongest operators improved their numbers by double digits, while the weakest saw theirs decline. Same tool. Opposite outcomes. The variable that separated them wasn't access to the technology. It was judgment.

That finding should reshape how executives think about the AI rollouts they are running right now.

The assumption embedded in most deployments is that the tool raises everyone. Hand a capable model to the team and the floor comes up. The research points the other direction. AI amplified the judgment already in the room. Where judgment was strong, the technology compounded it. Where judgment was thin, the technology accelerated the wrong decisions and the gap widened.

The mechanism is the part worth sitting with. When researchers looked at the data, the high and low performers were asking the AI similar questions and receiving similar answers. The difference was what each group did with the output. Strong operators discriminated. They recognized the recommendation that fit their specific situation and disregarded the generic ones. Weaker operators reached for the suggestions that sound universally reasonable, like cutting prices or spending more on advertising, the kind of advice that quietly erodes margin without addressing the actual constraint on the business. The model could not tell them which was which. Only judgment could do that.

This is the reality executives are walking into as they push AI across leaner teams. The dominant force in most organizations today is doing more with fewer people, and AI is the lever everyone is pulling to make that math work. But a compressed organization raises the cost of every decision. There is less slack to absorb a bad call and fewer people to catch it before it lands. AI does not reduce the demand for judgment in that environment. It raises it, because the technology now generates more plausible options, faster, than any leader can evaluate without a developed capacity to evaluate. The bottleneck moves from generating ideas to discriminating between them.

A second study makes the same point from a different angle. When AI was used as a sounding board, it helped less-experienced people think through a problem. When it was used as a ghostwriter that simply produced the work, it delivered no benefit to experts and actually degraded their output. The distinction is the whole argument. AI is useful as a thinking partner and corrosive as a substitute for thinking. The leaders who hand the judgment to the machine do not just get worse answers. They stop building the capability that produced good answers in the first place.

This is where the question of AI in the organization becomes a question of leadership development, not procurement.

Judgment of this kind is not a feature you license. It is developed. It comes from operating under pressure, owning the outcomes of your decisions, and building the internal capacity to read a situation accurately and act on it when the stakes are real. The AVEVV Framework describes the outer game, the observable disciplines a leader practices: setting direction, closing the gap between decision and delivery, holding the standard. Coaching addresses the inner game, the developed capacity beneath those behaviors, the part that determines whether a leader can tell a precise answer from a confident-sounding one when both arrive in the same response.

AI changes the outer game on a weekly basis. New tools, new capabilities, new ways to compress work. It does nothing for the inner game. An organization that invests heavily in the tools and not at all in the judgment to wield them is buying acceleration without steering. It will move faster in whatever direction its leaders point it, including the wrong one.

The competitive picture follows from there. Within any given market, leaders will soon be working with broadly the same models. The technology will not be the differentiator, because it will not be scarce. What remains scarce is the leadership capacity to use it well: to interrogate the output, to recognize when a recommendation is generic, to apply experience to a specific decision under real constraints. That capacity compounds. The organizations that build it will pull away from the ones that assumed the tool was enough.

The firms that win the AI transition will not be the ones with the best technology. Everyone will have the technology. They will be the ones whose leaders can tell a good answer from a confident one.