Most of the AI conversation in markets runs through hardware. How many GPUs did the hyperscalers buy, how big is the next data center, what did NVIDIA guide to this quarter. Those numbers are enormous and they matter. But they measure spending, not payoff. If you want to know whether all of this capital is actually producing something, the more honest scoreboard is showing up somewhere less glamorous: in the labor data. And what that data says is that AI is not simply adding or subtracting jobs. It is splitting the workforce into two tracks, and the gap between them is widening fast.
A workforce splitting in two
PwC’s 2026 Global AI Jobs Barometer, which combines analysis of close to a billion job ads across dozens of countries, lays out the divide plainly. On one side are jobs PwC calls “professionalised,” where AI takes over routine tasks and leaves humans to do the higher-judgment work. On the other are “democratised” jobs, where AI lowers the barrier to entry and spreads a skill more widely. The professionalised roles are growing roughly twice as fast as the democratised ones, and wages in those roles have grown about 42% faster since 2021.
The premium for knowing how to use these tools has gotten hard to ignore. PwC found that the wage premium for workers with AI skills reached about 62% this year, up from roughly 57% a year earlier. Jobs that require specific AI skills, things like machine learning or prompt design, are growing around eight times faster than the overall job market. Read that next to the headlines about layoffs and you get a more accurate picture. This is less a story of machines replacing people wholesale and more a story of a sorting mechanism, with rewards flowing to the workers and companies on the right side of it and pressure building on the rest.
The number an investor should actually watch
Here is where it connects to the market. For all the debate about whether AI earnings justify AI spending, there is a cleaner metric buried in this research: revenue per employee. PwC found that industries most exposed to AI saw roughly 3 times the growth in revenue per worker, about 27%, compared with around 9% for the least exposed industries. Labor productivity growth was roughly 40% higher at the most exposed companies than at the least exposed ones.
Why does that matter more than another capex headline? Because revenue per worker is where the rubber meets the road. A company can buy all the chips and cloud capacity it wants, but if that spending is real, it should eventually show up as more output per person, fatter margins, and the ability to grow the top line without growing headcount at the same pace. That is the thesis underneath the entire AI trade, stated in plain terms. So when you read the next round of earnings, the question worth asking is not just “how much did they spend on AI” but “is revenue per employee climbing, and is it climbing faster than at companies sitting on the sidelines.” If the productivity story is real, it will leave fingerprints there long before it shows up in a tidy headline.
A word of caution on the figures, though. These are correlations drawn from companies and industries that adopted AI early, and early adopters tend to be well run and well capitalized to begin with. Some of the gap is the tool. Some of it is simply that strong companies move first. Both things can be true at once, and a careful reader should hold that tension rather than assume the entire 27% is causal.
The entry-level squeeze
The uncomfortable part of the story sits at the bottom of the ladder. The first place AI seems to be biting is the work that used to belong to people just starting out. Recent labor data shows U.S. entry-level job postings down roughly 35% since early 2023, with the steepest cuts in fields like junior software development and data analysis, where some categories have fallen by more than half. Unemployment for recent college graduates aged 22 to 27 climbed to about 5.7% late last year, running above the rate for the workforce as a whole.
What is happening is subtle and worth understanding. The entry-level jobs that remain are changing shape. PwC found that entry-level roles most exposed to AI are now about seven times more likely to demand traditionally senior skills, things like leadership, judgment, and face-to-face persuasion. In other words, when AI absorbs the routine grunt work that junior employees used to cut their teeth on, the remaining junior roles quietly get “seniorised.” The first rung of the ladder does not disappear so much as rise out of reach for someone with no experience.
For the economy, that is a slow-burning problem. The routine tasks that AI now handles were also how a generation learned its trade. If you automate the apprenticeship, where does the next cohort of senior people come from? It is the kind of question that does not show up in a quarterly print but shapes the labor force a decade out, and it is one investors in human-capital-heavy businesses should be thinking about now rather than later.
Where the proof shows up
The cleanest test of whether any of this is real is whether companies will pay for it on a recurring basis. That is why the rise of so-called agentic AI, software that does not just answer questions but actually carries out multi-step tasks, is worth watching closely. Gartner expects roughly 40% of enterprise applications to embed task-specific AI agents by the end of this year, up from less than 5% in 2025. That is a steep curve.
The early commercial evidence is mixed, which is exactly why it is useful. Salesforce’s Agentforce, probably the most prominent pure-play example, had reached around $540 million in annual recurring revenue and roughly 18,500 customers by early this year, and the company has described it as its fastest growing product ever. Impressive, and yet still a small slice of a business that brings in more than $30 billion a year. The promise is large and the monetized reality is, so far, modest. That gap is the whole investment debate in miniature.
And not every project lands. Gartner has warned that more than 40% of agentic AI initiatives could be scrapped before the end of 2027, undone by unclear returns, rising costs, or the simple difficulty of bolting autonomous software onto messy real-world workflows. Surveys keep finding that a large majority of organizations report serious friction in their AI rollouts even as they pour money in. So the productivity gains are real and measurable in the aggregate, and at the same time plenty of individual efforts will fail. Both are true. The dispersion between winners and losers may end up being the defining feature of this cycle.
What does all of this add up to for someone trying to make sense of the market? Mostly a suggestion to change where you look. The capex numbers tell you how much conviction the spenders have. The labor and productivity data tells you whether that conviction is being vindicated on the ground. Revenue per worker, the AI wage premium, the health of entry-level hiring, the recurring revenue of the first real agent products: these are slower, quieter signals than a chip launch, but they are closer to the truth of whether this technology is paying for itself. The hardware story is about ambition. The jobs story is about results, and results are what eventually settle every argument.
_______________________________________________________________________________________________________
This article is written for educational and informational purposes only and does not constitute financial or legal advice. The views and analytical frameworks presented draw on publicly available information and reported commentary from industry participants. Readers are encouraged to consult primary sources and form their own informed views on these complex topics.
