Do people who choose to use an AI tool a lot shift their work differently than people randomly handed the same tool?

This explores whether people who chose to use Copilot heavily changed the kind of work they do differently from workers who were randomly assigned the tool. The short answer is that this collection doesn't contain that comparison. It does contain the randomized half, and that half explains why the comparison matters.

The randomized evidence comes from a six-month Microsoft trial across 66 firms and more than 7,000 knowledge workers Does Copilot change what knowledge workers actually do?. People who got Copilot saved about two hours a week on email. But the mix of their work did not change: they did the same amount and the same types of tasks as before. The authors' explanation is the part worth taking away. Email is something you do alone, so one person with a better tool can speed it up right away. Most other work depends on coordinating with a team. Changing that work requires the whole team to change how it operates, and that hadn't happened yet.

That explanation suggests what a self-selected comparison would probably show, though no note here tests it. Heavy users who sought out the tool would likely report bigger changes. Part of that would be because they were already the kind of people who reorganize their own work, which is exactly why studies assign the tool at random. The coordination argument also limits how far even enthusiasts could go. If a task depends on teammates who haven't changed their habits, one keen user can speed up their own solitary work but can't redesign the shared work alone. A power user's experience may describe what's possible for one person more than what changes for a whole organization.

So the takeaway is this: when you read a striking story about how AI "transformed" someone's job, ask whether that person chose the tool and whether their work depends on other people. The randomized result suggests that the first, easy gains go to solo tasks like email. Bigger changes to how work is divided up seem to wait for teams to adapt together. To see self-selected and assigned users compared directly, you'll need sources beyond this collection.