Something new is showing up when organizations measure their own people: the spread in output between employees doing the same job is widening, and the line that predicts which side of the spread someone lands on is increasingly whether they have genuinely worked AI tools into their day. Recent workplace surveys report heavy AI users saving on the order of a full working day's worth of hours per week, with leaders describing their strongest AI adopters as several times more productive than colleagues who have not engaged. The precise multiples deserve skepticism - self-reported productivity always does - but the direction is consistent across too many studies to dismiss: inside the same company, on the same tasks, a real gap has opened between the people who use these tools well and the people who do not use them at all.
What makes this a management issue rather than a curiosity is that the gap is not primarily about talent. The same person is meaningfully faster at drafting, summarising, analysing and preparing when routine cognitive work is delegated to a tool than when it is not - so the divide sorts people by adoption, not ability. And adoption, left to itself, follows temperament: the curious experiment, the busy defer, the skeptical abstain, and the anxious avoid the subject entirely. A company that does nothing is effectively letting personality decide which of its employees get a large productivity assist, then measuring everyone as if the race were even. That is unfair in both directions - it flatters the adopters' rankings and quietly punishes solid performers whose only shortfall is not having found three hours to learn a tool.
There is also a subtler dynamic: much of the adoption is invisible. Employees frequently use AI without telling anyone, either because no policy exists, or because the policy is unclear, or because they suspect that revealing the time saved will simply raise their targets. The result is a company that does not actually know where its productivity is coming from, cannot spread the practices that work, cannot check whether sensitive material is being pasted into unapproved tools, and cannot distinguish an employee who produces more from an employee who produces more with help on tasks where the help matters. Silence around AI use is not neutrality; it is a guarantee that the gap grows unevenly and unexaminedly.
The constructive response is to treat AI proficiency the way businesses treat any capability that lifts output: make it explicit, teach it, and normalise it. Find the internal heavy users and have them show their actual workflows - concrete demonstrations move colleagues in a way mandates do not. Give the non-adopters sanctioned time and a starting task rather than exhortation. Set clear rules about what may and may not go into which tools, so honest use is safe to disclose. And resist the temptation to immediately harvest every saved hour, because employees who see the benefit confiscated will stop reporting it. The gap between AI users and everyone else is now real enough to show up in the numbers. The only question a business controls is whether that gap becomes a shared capability or a quiet, growing unfairness.