Does AI at work stay a solo habit — one person drafting or coding alone — or does it actually spread across teams?

This explores whether AI use at work stays confined to narrow, individual tasks (one person drafting, documenting or coding alone) or spreads across how organizations work together. The corpus has no direct measurement of where adoption lands. It has nothing like a task-by-task survey of knowledge workers. What it does have are several separate lines of evidence that all point the same way: AI spreads easily to individuals and stalls when it reaches the organization.

The clearest statement comes from Does easier tool-building actually solve enterprise adoption problems?. Evans argues that the limit isn't technical. Most workers don't see their own tasks as something that could be automated, and real enterprise adoption needs decisions that cross departments and take months or years. Making tools easier to build lowers the cost for one person to try something, but it does nothing about these two barriers. The likely result is solo, self-directed use: people automate what they can see and control, such as writing, summarizing and documenting, while shared workflows stay as they were.

Developer data shows a different kind of narrowing. In Why do developers keep using AI tools they don't trust?, AI use reaches 80% of developers while trust in its accuracy falls from 40% to 29%. Adoption is broad, but the work changes shape: code that looks correct but contains subtle errors adds a checking burden. So broad uptake doesn't necessarily mean AI is doing broad work. It can mean many people doing the same narrow loop of generating and then verifying. Part of the reason is that AI's working context is unstable. It shifts with every prompt and retrieval (How does AI context differ from conventional software context?), which makes it hard to hand off between people or build into a team process.

The less obvious angle concerns what kind of work AI can take on alone. Can one AI system complete a full research cycle end-to-end? shows one system going from idea to a workshop-accepted paper on its own. That suggests AI can handle a complete solo knowledge cycle, not just pieces of one. Yet Can AI ever gain expert community trust through participation? argues that expert authority comes from taking part in a community and building a track record over time, which AI cannot do. Put together, these suggest a ceiling. AI may get very good at the individual production side of knowledge work (drafting, experimenting, documenting), while the social side, where work gets judged, trusted and adopted, stays with people.

So the honest answer is that the corpus points toward AI being adopted widely by individuals but not deeply by organizations. That isn't proven, and "narrowing toward documentation" specifically isn't something any source here measures. If you want to go further, start with Evans for the organizational argument and the Stack Overflow survey for what happens to trust when use outpaces it.