The Productivity Trap: Why Doing More Won't Make Experienced Workers Wealthier

The 45-year lie inside “just be more productive”

Since 1979, net productivity in the US, output per hour worked, less depreciation, has grown 90.2%. Over the same period, the hourly pay of a typical production or nonsupervisory worker in the private sector has grown 33.0%, according to the Economic Policy Institute’s long-running productivity-pay series. Productivity has grown roughly eight times faster than typical pay across that stretch. If pay had simply tracked productivity, the way it did for the three decades before 1979, EPI calculates the typical worker would be earning $16.40 more per hour today, $13.53 of it in wages and the rest in benefits, not as a windfall, just as the outcome of the same relationship holding.

That’s the baseline “work harder, get ahead” was quietly built on top of for nearly half a century, and it stopped being true for most workers a long time before AI showed up. The advice never updated to match. What AI changes isn’t the existence of the gap. It’s the speed at which the mechanism that produces it is now operating, and how visible that mechanism has become in a single year’s data.

Free tool

Enterprise AI Value & Adoption Dashboard

Check where your own AI programme’s gains are actually landing, cost savings and margin, or revenue and capability your people get paid more to run.

→

AI isn’t a neutral productivity tool. It’s built to be capital-biased

Daron Acemoglu’s 2025 paper, “The Simple Macroeconomics of AI,” is the most rigorous attempt yet to size what AI actually does to growth and distribution, using a task-based model rather than speculative scenarios. His headline growth number is deliberately modest: no more than a 0.66% increase in total factor productivity over ten years from currently exposed tasks, revised down to under 0.53% once harder-to-learn, context-dependent tasks are accounted for. That number alone punctured a lot of the more excitable 2024 and 2025 forecasting.

The finding that matters more for this piece is the distributional one, and it’s stated plainly: AI is predicted to widen the gap between capital and labor income. Even in scenarios where AI’s productivity boost to low-skill tasks helps some workers, the paper finds no evidence that AI reduces labor income inequality overall, and it shows AI’s aggregate effect skewing toward capital rather than labor, independent of how fast or slow the total growth number turns out to be. In other words, the size of the AI productivity dividend is still genuinely uncertain. Who captures it is not.

That prediction isn’t waiting for confirmation.

The causes aren’t singular. Weakened union bargaining power, a federal minimum wage frozen since 2009, and tax policy that favours capital gains over wages have all been building this for decades. AI is not the sole author of that trend. It’s the newest and fastest-moving contributor to a mechanism that was already well established, at exactly the point Acemoglu’s model says it should be accelerating, not slowing, the divergence.

Follow where the AI productivity gains are actually going

If AI-driven productivity were flowing to the workers producing it, the 2026 numbers underneath the labor-share figures would look different. They don’t. S&P 500 companies spent a record $1.02 trillion on share buybacks in the twelve months to Q3 2025, up from $918.4 billion the year before, according to S&P Dow Jones Indices. That’s capital being returned to shareholders at the same time employers are running the fastest pace of AI-attributed layoffs on record: Challenger, Gray & Christmas has tracked 101,743 US job cuts citing AI in 2026 through June, 23% of every announced cut, already well ahead of the 54,836 recorded across all of 2025 and part of 173,568 cumulative AI-cited cuts since the firm started tracking the category in 2023.

Wealth concentration is moving in the same direction at the top of the distribution. Oxfam’s January 2026 inequality report found global billionaire wealth rose more than 16% in 2025 to a record $18.3 trillion, three times faster than the average annual pace of the previous five years, with the number of billionaires passing 3,000 for the first time. None of this proves AI alone is driving billionaire wealth growth. It does show the same pattern repeating at every scale the data can measure this year: gains concentrating with owners of capital, not with the workforce whose output is generating them.

Where the value is measured	2026 direction	What it means for a worker’s pay
Labor share of corporate income	Fell to 71.3%, down from 77.8% in 2020	A shrinking share of every dollar of output reaches paychecks, even when output rises
S&P 500 buybacks	Record $1.02 trillion, 12 months to Q3 2025	Capital returned to shareholders faster than it’s reinvested in headcount or wages
AI-cited job cuts	101,743 through June 2026, 23% of all cuts	Productivity gains from AI are being realised as cost reduction, not redistributed as pay
Billionaire wealth	Up 16%+ in 2025 to $18.3 trillion, 3x the 5-year average pace	Ownership of capital, not labor, is where this year’s fastest gains are landing

Four independent data sources, four different methodologies, the same direction in the same year. That’s what makes 2026 different from any single one of these numbers on its own.

What this actually means if you’re experienced, trained and ready for the market

The workers currently best insulated from AI-driven displacement are the ones with deep, tacit expertise, the judgment AI still can’t reliably replicate. That’s real protection, and it’s worth naming plainly rather than pretending experience counts for nothing. But protection from being replaced is a different claim from participation in the value being created, and conflating the two is where a lot of “reskill and you’ll be fine” advice quietly breaks down.

Acemoglu, together with David Autor and Simon Johnson, published a framework in February 2026, “Building Pro-Worker Artificial Intelligence,” through the Hamilton Project at Brookings, that names exactly why. They split technological change into five categories by what it does to a worker’s bargaining position, not just their employment status.

Category	What it does	Effect on the worker
Labor-augmenting	Helps a worker do their current tasks faster	Neutral to modestly positive, depends on who captures the time saved
Capital-augmenting	Increases the value of capital relative to labor	Negative: raises the return to ownership without raising the return to work
Automating	Transfers a task from a worker to a machine	Negative for the worker performing that task
Expertise-leveling	Lets less-experienced people do specialist work	Ambiguous: can raise output while reducing the market’s willingness to pay for scarcity
New task-creating	Expands the range of valuable work humans can do	The only category the authors call unambiguously pro-worker

Most enterprise AI deployment today sits in the automating and expertise-leveling rows, not the new-task-creating one, and that placement is a decision made in a business case, not a property of AI itself.

Expertise-leveling is the row that applies most directly to an experienced professional. It doesn’t take your job. It takes the scarcity premium the market used to pay for your judgment, by making a version of that judgment available to people who didn’t spend twenty years building it. Your output can stay valuable to your employer at the exact same moment the market’s willingness to pay a growing wage for it flattens, because the thing that made your judgment scarce is being deliberately, and often successfully, engineered away. That’s a subtler failure mode than a layoff notice, and it’s much harder for an individual worker to see happening to them in real time, because their paycheck doesn’t stop, it just stops growing relative to what they’re actually producing.

Free tool

Outcome as a Service Qualification Checklist

If your delivery time is shrinking because of AI, this checks whether your pricing is still tied to hours, or to the outcome that’s actually worth paying for.

→

What rethinking business and wealth development actually requires

None of this is an argument against deploying AI, and it isn’t an argument that experienced workers are powerless. It’s an argument that the old assumption, that being productive and being paid for it are the same act, needs to be replaced deliberately, at both the business and the individual level, because the data says it’s already coming apart on its own.

For businesses, that starts with using the five-category framework above as a genuine screen on the AI investment case, not a compliance checkbox: ask whether a given deployment is automating and expertise-leveling your way to a lower cost base, or genuinely new-task-creating in a way that expands what your people can be paid to do. Then close the loop that data alone won’t close: build the mechanisms, profit-share tied to AI-driven margin expansion, equity for the people who built or ran the system, outcome-based contracts that price the result instead of the hours it used to take, that route some of the captured value back to labor instead of defaulting it to buybacks by omission. Acemoglu, Autor and Johnson’s own policy recommendations, tax incentives favouring labor, stronger worker voice mechanisms, targeted public investment in worker-complementing tools, are aimed at exactly this gap, because the market alone hasn’t been closing it for 45 years and there’s no evidence it starts closing it now.

For experienced individuals, the strategic shift is from optimising for productivity to optimising for a stake in the outcome. Being fast and good at your job was never, on its own, the thing that reliably built wealth, the 1979 to 2025 data settles that. What builds wealth is owning a piece of the value your work creates: equity, profit share, a business model priced on results rather than time, assets you control rather than hours you sell. AI proficiency is becoming table stakes, not a differentiator, the same levelling effect already visible in how fast fluent, competent output has gone from a rare skill to a free one. The scarce thing left to negotiate for isn’t skill. It’s a structural claim on the outcome that skill produces.

The banks and law firms cutting entry-level classes while running record AI-cited layoffs and record buybacks in the same fiscal year aren’t behaving irrationally. They’re behaving exactly as the incentives in front of them predict. Changing what “productive” buys a worker in 2027 and beyond isn’t a matter of individuals working harder or getting more AI-fluent. It’s a matter of whether enterprises deliberately build the mechanisms that route value back to the people creating it, before the gap that’s been widening since 1979 becomes the permanent shape of the AI-era economy instead of an inherited flaw inside it.

---

The AI Governance & ROI Executive Programme works through exactly this question at the organisational level, how to structure AI investment and compensation so productivity gains route back to the people creating them, before the gap shows up in next year’s attrition numbers instead of this year’s plan. Details are on the workshops page.