Does AI at work mostly boost a small group of power users, or lift everyone's output?

This explores whether the benefits of AI at work go mainly to a small group of heavy, skilled users or reach workers broadly. The corpus leans toward concentration, along several lines at once. Where workers have actually handed tasks over to AI in structured workflows, those tasks cluster in information-heavy jobs. That spread follows what the technology can do, not how popular chatbots are, and it doesn't match older predictions that routine work would be automated first (Where have workers actually delegated tasks to AI?). A survey of executives finds the same thing at the firm level: measurable gains gather in high-skill services and finance (Do AI productivity gains feel larger than they actually measure?). Across countries, Claude usage rises with GDP per capita. Wealthier nations use it for many different things, while poorer nations mostly use it for coding (Does AI adoption follow wealth and mature over time?).

The less obvious finding is why it concentrates. It isn't just about who has access. Productivity gains show up when people use AI on work they already know how to do. When workers used AI to learn a new skill, the gains disappeared and their learning suffered (When does AI actually boost worker productivity?). So AI multiplies expertise you already have more than it builds new expertise, which pushes the advantage toward people who started ahead. A related argument says AI mainly speeds up the middle 'doing' part of knowledge work. Deciding what to do and delivering the result stay human and may even grow (Does AI really compress all layers of knowledge work equally?). The people who benefit most are those who already have the judgment to direct and check AI's output.

Individual gains and collective gains can also pull in opposite directions. Scientists who use AI publish about 3× more papers and collect about 4.8× more citations. Yet across science as a whole, the range of topics studied shrinks and collaboration falls by 22%, because work drifts toward problems that already have lots of data (Does AI help individual scientists while narrowing scientific focus?). Power users can win while the shared field gets narrower. Microsoft's research argues that the next frontier is team-level productivity, not individual tools, but it offers this as a design goal without evidence that it has happened yet (Can AI boost how teams work together?).

There is also a measurement problem. In one interview study, 86% of general workers said AI saved them time, yet about 70% hid or played down their use because of stigma at work (Why do workers hide productivity gains from AI use?). Executives also report bigger gains than the numbers show (Do AI productivity gains feel larger than they actually measure?). Use may be more widespread than official figures suggest, but the gains are hard to see and unevenly credited. Over time, usage also seems to shift from handing off whole tasks toward working alongside AI and learning from it (Does AI adoption follow wealth and mature over time?). That suggests broader benefits may come with experience, not just with access.

One gap in the corpus: it has no direct within-company comparison of novice and expert workers using the same tool. The evidence here is about sectors, countries and skill conditions. Read it as 'gains follow existing expertise and wealth', not as a final verdict on how gains are split inside any one workplace.