When people lean heavily on generative AI, does their work tilt toward solo documents and away from colleagues?

This explores whether generative AI changes the mix of work people do, not just their speed. The most direct answer in the corpus is yes, and the direction is surprising: heavy AI users move toward working alone. In one study of workplace software logs, heavy generative AI users increased their actions in productivity apps (documents, spreadsheets) by 21.2 percent, but their communication actions grew by only 7.1 percent. The work rebalanced toward solo documentation and away from team coordination Does generative AI shift knowledge workers away from communication?. A field experiment at Procter & Gamble helps explain why. Individuals using AI produced solutions as strong as two-person teams without AI, and their ideas were more balanced across professional backgrounds, as if the AI supplied the missing colleague's perspective Can generative AI replace the benefits of having a human teammate?. Put the two findings together and a hypothesis emerges: some of the messaging, meetings and back-and-forth that used to be how knowledge got combined may be getting absorbed into a private conversation with a model.

The shift is real but uneven, and in many places it hasn't started. At Argonne National Laboratory, a survey and interviews of 66 staff found that use of an internal chatbot stayed small and experimental. It was concentrated in structured writing tasks, and few people had built it into everyday workflows How are national lab staff actually using generative AI?. So the first tasks to move are the ones that look like drafting. The deeper reorganization of work is still mostly potential.

Another shift is less visible: who can do which tasks. In a randomized experiment with 1,174 adults, AI closed about three-quarters of the performance gap between more- and less-educated participants on a business problem-solving task Can AI narrow the education performance gap?. That suggests AI opens up tasks that used to require credentials. But a separate finding complicates this: workers who did better with AI showed no improvement when they later did similar tasks on their own Does AI assistance help workers learn lasting skills?. Workers may be moved into new tasks without picking up the skills those tasks used to build. Whether that widens or narrows inequality depends on how organizations deploy the tools, not on the technology itself Does generative AI inevitably worsen or reduce inequality?.

What's left for the human? Examples from beyond office work suggest the job moves from producing work to checking it. In political advertising, AI can write and pre-test personality-tailored ads without human writers, so the constraint shifts from writer time to compute cost Can generative AI scale personality-targeted political persuasion?. In research, frontier agents mostly recombine known techniques rather than invent new ones Do frontier AI agents actually conduct novel research or just optimize?. Deep research agents sometimes invent sources and evidence to look rigorous Why do deep research agents fabricate scholarly content?. Both findings mean someone still has to judge novelty and verify claims. So one plausible answer to the question is that AI pushes knowledge workers toward solo drafting and toward checking machine output, and away from coordinating with each other. A caveat: the corpus has only one direct measurement of the task-mix shift, so treat that conclusion as a well-supported hypothesis, not a settled fact.