Source: Caplin et al. (2025) — “The ABCs of Who Benefits from Working with AI,” Management Science
The most tempting sentence in AI is: “AI makes people more productive.”
It sounds clean. It sounds measurable. It sounds like something that belongs in a board deck with a blue gradient.
But the more interesting question is: which people?
This paper looks at who actually benefits from working with AI. I like that framing because it refuses to treat “workers” as one big identical blob. People come into AI collaboration with different skills, confidence levels, habits, and weaknesses. Naturally, AI does not land on everyone the same way.
That feels obvious once you say it, but organisations often skip this step.
They roll out one tool, one training session, one cheerful email from leadership, and then act surprised when usage patterns are messy. Some people become dramatically faster. Some barely change. Some trust the output too much. Some refuse to touch it because they had one bad answer and now the entire technology is personally offensive.
The useful lesson is that AI benefits are distributed.
A tool can help lower-performing workers by giving them structure. It can help high-performing workers by speeding up routine work. It can also create very different outcomes depending on whether the person knows how to ask, check, revise, and decide.
So the question is not simply whether AI works.
The question is who it works for, under what conditions, and what support they need to use it well.
My takeaway: AI transformation should not be designed for an imaginary average employee. That person does not exist. The real work is understanding the different groups of users, then building training and workflows around them. Less glamorous than a launch announcement, but much more useful.