The chief operating officer ran the meeting from a second screen. Someone asked for the prior quarter's exception report split out by region, and before the question was finished she had it drafted in a chat window, dropped into the deck, and moved on. She had folded the tool into her day so completely that she had stopped noticing she was using it. Afterward, walking out, she said the thing I now hear in some version most weeks. "I don't get why my people won't touch it. I'm on it constantly."

She was right about herself and wrong about what that told her. Her ease was real. The conclusion she drew from it, that adoption was basically handled and the holdouts were a motivation problem, is where the trouble started. The people who run organizations have picked up these tools faster than anyone else in the building, and that head start is the very thing that hides the floor below them.

Fluency at the top is the worst vantage point

BCG ran its annual survey of more than ten thousand leaders, managers, and frontline employees across eleven countries last year. More than three-quarters of leaders and managers said they use generative AI several times a week. Among frontline workers, regular use had stalled at 51 percent.1 Read that gap slowly. The half of the workforce closest to the actual work, the people whose output the whole AI case rests on, are the half that has not moved.

The same survey found the lever that moves them, and it is not a better model or a wider pool of licenses. When leaders visibly back the tool, the share of frontline employees who feel positive about using it rises from 15 percent to 55 percent. Only about a quarter of those employees say they get that kind of support.1 The input that would close the gap is the cheapest one in the room. It is also the one a fluent leader is least likely to think she needs to provide, because from where she sits the problem already looks solved.

What the leader cannot feel

There is a second effect that almost nobody at the top accounts for. When a manager is visibly skilled with a tool and talks about it often, the people underneath do not always hear encouragement. Some of them hear a standard they have not been shown how to meet, and a question about whether their job survives the year. The leader experiences her own use as freedom. The person three levels down can experience the same enthusiasm as a kind of watching. So they wait. They let the tool sit there, and they get very good at looking busy with the work they already know how to do.

This is why the "why won't they use it" framing keeps failing. It treats a condition of the system as a flaw in the person. The more useful question turns the leader from frustrated spectator into the one who can actually change something. Not why aren't they adopting this, but what would have to be true for this to feel safe and worth their time. That is a different conversation, and it starts with the leader admitting that the one perspective she trusts most, her own, is the one she should trust least here.

The executive who uses AI every day is the worst-positioned person in the company to judge whether it is working everywhere else. The fluency that makes the tool vanish into her day also makes the adoption gap vanish from her view.

The bill comes due downstream

While the top of the house assumes the rollout is on track, the numbers underneath tell a colder story. MIT's Project NANDA studied enterprise AI last year and found that about 95 percent of organizations were seeing no measurable return on their generative AI spending.2 Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027, citing unclear value and rising costs as the reasons.3 These are not stories about models that failed. They are stories about adoption that never reached the people who would have turned a pilot into a result, while the budget kept flowing toward the next tool.

The spending pattern follows the blind spot. More licenses, another platform, because those are the moves that feel like progress to someone who has already crossed the gap herself. The unglamorous input, a leader changing how she shows up around the work so the people below feel safe doing it badly at first, does not appear on any purchase order. It compounds slowly and it is hard to point to in a board update. It is also the only thing the evidence says actually moves the floor.

So here is the question for anyone who reads their own AI fluency as proof the organization has adopted. When was the last time you watched someone four levels down try to use one of these tools, and stayed quiet long enough to see exactly where they got stuck? If you cannot remember, the adoption problem you have is not one you can see from your desk. It is one your own competence has trained you not to notice.