When AI makes shopping, deals, or coordination frictionless, does that power spread out to regular people, or just pool with whoever owns the tools?

This explores whether AI's promise of removing friction (cheaper search, instant transactions, less busywork) actually spreads power around, or quietly pools it with whoever owns the tools. The short version from the corpus: friction rarely just disappears. It moves somewhere else, and who gains power depends on where it lands. No note in the collection directly tests whether friction reduction redistributes power, so treat what follows as a pattern pieced together from several angles, not a settled finding.

The clearest case comes from markets. The textbook expectation is that cheaper search pushes markets toward perfect competition, where no one has much pricing power. Bichler's model finds that AI shopping agents do slash technical search costs, but they bring in new frictions: agents have to guess what you want, information gets distorted along the way, and too many sellers pile in Do AI agents actually reduce friction in electronic markets?. So the power question shifts. It's less about who can afford to search and more about whose agent infers preferences well and who shapes the information the agent sees. Banking shows the same pattern from the opposite direction. If AI lets depositors move money in hours instead of days, that's real power handed to individuals. But it also means a weak bank can collapse faster, and Roche argues this risk is narrow and specific to individual banks, not system-wide Could AI agents speed up runs on individual weak banks?. Removing friction gives individuals power, and it also removes the buffer that protected fragile institutions.

At work, the deciding factor is what kind of friction gets removed. Brynjolfsson's distinction is useful here. AI that automates labor shifts bargaining power to whoever owns the technology, while AI that augments labor leaves workers with leverage Does automating labor concentrate power more than augmenting it?. Larsen-Ledet makes a finer point inside that distinction. Some friction is pointless, like re-explaining context the system should already know. Other friction is productive, because putting your work into words sharpens how you think about it Can generative AI reduce overhead work that adds no value?. Cutting the first kind strengthens workers. Cutting the second kind may weaken them. The labor data adds one more condition. When AI touches only a few tasks in a job, workers can shift to the tasks that are left and employment holds up. When it touches many, that escape route narrows Does concentrated AI exposure enable workers to adapt and reallocate?.

The less obvious part, and the one worth knowing, is that some friction was quietly holding up human power all along. The gradual-disempowerment argument says institutions stay aligned with people partly because they need people to do the work. That dependency is a kind of friction, and once it's automated away, ordinary people lose leverage they never had to claim explicitly Does incremental AI replacement erode human influence over society?. A related note points out that the goals AI systems actually get trained on are the ones that can be measured and added up, and those tend to leave out people's ability to say no Do welfare goals that prevent veto gaps actually exist in practice?. That's why some researchers suggest adding friction back on purpose: slowing development as a safety buffer Does slowing AI development actually prevent system failures?, or designing small frictions into collaborative tools. The evidence for deliberate friction is thin, though. The micro-friction proposal, for example, has never actually been tested Can micro-frictions boost rivalry without harming collaboration?.

So can friction reduction redistribute power? Yes, when it removes friction that protected the powerful, like search costs for consumers, slow withdrawals for depositors, or busywork for workers. It concentrates power when it removes friction that protected everyone else, like the need for human labor, a productive struggle, or a pause that leaves room for a veto. The useful question to ask about any 'frictionless' AI product is whose leverage that friction was holding up.