The thing AI pays for is the judgment it cannot do itself, and the data that looks like a story about tool fluency is really a story about that. PwC has read more than a billion job ads across 27 countries, and the picture it published this month is stark: jobs that ask for AI skills are growing about 69% a year, roughly eight times the 9% the wider market is growing, and they pay a wage premium of 62%, up from 57% the year before. The obvious conclusion is that the people who learned the tools fastest are the ones being rewarded, and so the move is to learn the tools. That conclusion is half right and pointed in the wrong direction. The roles pulling away are the ones where AI took the mechanical part of the work and left a person owning the part that takes a call, and the firms pulling away rebuilt the work so that is exactly what happened.
What the numbers say, and what they quietly settle
The headline figures are large enough to read as noise, so it is worth holding them still. AI-skill roles are growing about eight times faster than the overall labour market. The wage premium attached to those skills sits at 62%, up from 57% a year earlier, so the gap is widening even as the skills spread. The firms most exposed to AI, the top fifth by exposure, posted labour-productivity growth of 163% relative to 2018, close to five times the average among AI-exposed companies. None of that, on its own, tells you whether the money is flowing to people who can run a prompt or to people who can decide what the prompt is for. The detail that settles it is buried lower in the report, and it is the one worth carrying out of the building.
The premium is on judgment, not on the tool
AI-exposed entry-level roles are now seven times more likely to require senior-level skills such as judgment and leadership than they were before. Read that against the wage premium and the shape of the thing comes clear. A junior posting that used to ask for raw output now asks for the ability to weigh, to decide, to take responsibility for a call, because the raw output is the part AI already covers. The honest caveat is that the 62% figure is attached to AI skills broadly, which still includes tool fluency, so no one can claim the market is paying for judgment alone. What the market is plainly doing is lifting the floor on what a person has to bring once the tool is sitting next to them. The mechanical layer got cheap, and the price moved onto the layer above it, the one a model still cannot occupy.
This is why the firms at the top of the exposure ranking are seeing productivity gains that dwarf the average. A company that simply hands its people better tools gets a faster version of the same work, whereas a company that moves its people up to the judgment-heavy part and lets AI carry the rest changes what an hour of that person's time is worth, and the productivity line follows.
Two tracks are opening, and one is a choice
The report draws a line through the middle of the labour market, and the two sides are diverging on purpose. On one side sit what it calls professionalised roles, jobs like radiologists and recruiters where AI deepened the work and raised the bar for the human in it. Those roles are seeing twice the job growth and 42% faster salary growth than the roles on the other side. On the other side sit democratised roles, where AI made the task easy enough that almost anyone can do it, which is the same as saying the person doing it became interchangeable. The first track compounds, the second commoditises, and the difference between them is rarely the technology so much as whether the work was rebuilt so the person owns something AI cannot supply.
Which track a job lands on is decided long before the wage shows up. A leadership team that buys AI and asks its people to push more work through the same roles is building the democratised track without meaning to, because faster output on an unchanged job is the definition of making the person replaceable. A team that takes the same technology and redraws the role around the part only a person can hold is building the professionalised track. Same tools, same year, opposite outcome for the people inside.
This is rebuild the work, at the level of a job
The instinct most companies follow is to bolt AI onto the jobs they already have. The model gets added as a layer, the work stays the same shape, and the only visible change is that each person is expected to produce more. That is the track where the human gets cheaper, and the PwC data is what it looks like at scale across a billion postings. Rebuilding the work means something narrower and harder: taking a single role apart, handing the mechanical, repeatable, judgment-light part to AI, and rebuilding the job around the decisions, the trade-offs, and the calls that need a person to own them. The output goes up either way. The difference is whether the person doing it became more valuable or more interchangeable in the process, and that difference is the whole game.
This is the reason AI done well makes people worth more rather than less. When a role is rebuilt so its centre of gravity is judgment, the technology lifts the person while the firm captures the productivity gain at the same time. The companies posting 163% productivity growth since 2018 did not get there by asking their people to type faster. They got there by changing what their people spend their hours on, and letting the rest fall to the machine.
The uncomfortable question worth sitting with
In your company, are the people closest to AI being freed to do more judgment, or just asked to produce more output faster? The answer decides which of PwC's two tracks your roles are quietly drifting onto, and almost no one has been asked it out loud. If the honest answer is more output, the wage premium and the productivity gains in the data are accruing somewhere other than your business, to companies that used the same tools to move their people up while yours worked them harder. The technology is the same on both tracks. The judgment to rebuild the work around the part AI cannot do is the part a model never arrives carrying, and it is the part the market has started paying for.