The gains are real. Who gets them?

The argument about whether AI produces real productivity gains is quietly ending. Task-level studies keep landing in the same range, and analyses of augmentation-led adoption now put the prize in the trillions: research from Pearson presented at the World Economic Forum values upskilling-and-augmenting, rather than replacing, at between $4.8 and $6.6 trillion for the US economy alone. The interesting question has moved on. The gains are real; who gets them?

In most organisations, nobody has decided. The gains arrive diffusely, as time saved here and headcount avoided there, and are absorbed invisibly into margins and workloads. Meanwhile the people producing them draw their own conclusions. Workforce surveys make uncomfortable reading: around seven in ten US workers familiar with AI told EY they were concerned about it, and ManpowerGroup's 2026 Global Talent Barometer found technology confidence falling sharply even as regular AI use jumped. People are not resisting the technology. They are resisting a deal in which they carry the disruption and someone else keeps the dividend.

The employers experimenting with the answer

A small but growing set of employers is answering the question out loud. The most visible experiments involve time: four-day weeks at full pay, positioned explicitly as AI's dividend returned to staff, an idea even OpenAI has urged companies to trial. The earlier UK four-day-week trial remains the reference point: across 61 companies, 92% kept the shorter week after the pilot, with revenue broadly stable and far fewer staff leaving. Time is not the only currency: better pay, funded learning, and simply more interesting work are all versions of the same decision.

We are not arguing every organisation should adopt a four-day week. We are arguing that the allocation of AI gains is a decision, that it is currently being made by default almost everywhere, and that the default is corrosive.

Why this is an adoption question

Here is the mechanism that makes this commercially interesting rather than merely fair. Adoption is the binding constraint on AI value; the majority of failed programmes fail because people never really used the thing. And people adopt what benefits them. A workforce that can see a credible line from "this system works" to "my work gets better" will make the system work, tune it, defend it, and teach it to the next hire. A workforce that cannot see that line will comply, minimally, until the programme joins the others in the drawer.

So the gain-sharing decision is not a generosity tacked onto an AI strategy. It is load-bearing. It converts the people closest to the work from the programme's audience into its engine, and it is far cheaper than the alternative, which is buying the technology twice: once from the vendor, and once more in the slow grind of pushing it onto people who have no reason to want it.

Making the decision decidable

The practical obstacle is that most organisations cannot share what they cannot see. If the gains are unmeasured, every conversation about allocating them collapses into assertion. That is why, in our delivery and operating work, we insist on measuring value in hard terms: cost per outcome, hours returned, decisions improved. Measurement is usually sold as governance. It is also what makes generosity possible: you can only give people a share of a number you actually have.

The companies that win with AI are the ones whose people want it to win. Wanting is not manufactured by communications plans. It is purchased, honestly, with a visible share of the gains.