The Economy Grew 2.2% Last Year. It Created Fewer Jobs Than a Single Amazon Warehouse.
Labor's share of GDP just hit the lowest point in 78 years of measurement. Employment intensity of growth has collapsed 90% since the 1960s. The BLS overcounted 2025 job creation by 3.2×. The Great Decoupling isn't a prediction. It's the present.
One hundred and eighty-one thousand.
That's how many jobs the U.S. economy created in all of 2025. Not per month — total. For the entire year. On a labor force of 158 million.
You probably remember a different number. Throughout 2025, the Bureau of Labor Statistics reported job creation of 584,000 — a figure the Federal Reserve used to set interest rates, politicians cited to claim a "strong labor market," and financial media used to dismiss concerns about AI displacement. Then, in February 2026, the BLS quietly published its annual benchmark revision. The real number was 181,000. The economy had created 3.2 times fewer jobs than reported.
Meanwhile, GDP grew 2.2%. Corporate profits hit a record $1.87 trillion. Productivity surged 4.9% in Q3. By every measure of economic output, 2025 was a good year.
For the economy. Not for workers.
The Number That Changes Everything
Economists have a term for it: the Great Decoupling. Brynjolfsson and McAfee named it in 2013, tracking the growing gap between productivity and employment. But what we're seeing now makes their original observation look quaint.
Here's the trajectory of American jobless recoveries — except the current one isn't a recovery at all:
| Recession | GDP Recovery | Employment Recovery |
|---|---|---|
| 1948-49 | 4 quarters | 3 months |
| 1957-58 | 3 quarters | ~8 months |
| 1990-91 (first "jobless recovery") | 3 quarters | 26 months |
| 2001 | 2 quarters | 39 months |
| 2007-09 | 8 quarters | 76 months |
| 2025-26 | GDP still growing | Employment negative |
Read the bottom row again. Every previous episode involved a recession — GDP fell, then recovered, and employment eventually followed. 2025-26 is categorically different. GDP never fell. The economy is growing. Employment is stagnant or declining. There's no recession to "recover" from.
This isn't a jobless recovery. It's jobless growth. And the data says it's becoming permanent.
The 78-Year Low
The most structural indicator isn't a jobs number. It's a share.
Labor's share of GDP — the portion of economic output that goes to workers as wages, salaries, and benefits — hit 53.8% in Q3 2025. That's the lowest since the Bureau of Labor Statistics began tracking it in 1947.
For half a century, labor share was so stable that economists treated it as a constant — a fundamental feature of capitalism that hovered between 57% and 62%. Textbooks printed it as an axiom. Then it started falling:
| Period | Labor Share of GDP |
|---|---|
| 1947-2000 | 57-62% (stable) |
| 2001-2015 | Declined to ~56% |
| 2016-2019 | Recovered slightly to ~57% |
| Q3 2025 | 53.8% (record low) |
"That decline in the share of labor has got to be either falling earnings or falling numbers of people," Raymond Robertson of Texas A&M told Fortune. "The falling share of income is having to do with the shift towards capital."
Translation: the economy is producing more. Workers are getting less of it. And the gap just hit a level we've never measured before.
Employment Intensity: The Ratio Nobody Talks About
There's a quieter metric that tells the structural story even more clearly. Employment intensity is the ratio of employment growth to GDP growth — how many jobs each point of economic expansion creates.
In the 1960s, one point of GDP growth generated roughly 0.5 to 0.7 points of employment growth. By the 2010s, the ratio had fallen to 0.2 to 0.3. For 2025: GDP grew 2.2%. Employment grew 0.1%.
That's an employment intensity of ~0.05. One-tenth of the 1960s level. A 90% collapse in the economy's ability to convert growth into jobs.
This decline isn't explained by the business cycle. It crosses expansions and recessions alike. It reflects four compounding waves, each of which eliminated the escape route from the previous one:
Wave 1: Manufacturing (1979-2010). Output roughly doubled while employment fell by a third. Policy response: retrain displaced factory workers into service jobs. Result: opioid crisis, deaths of despair, political radicalization in deindustrialized counties that, according to Frey, Berger, and Chen (2018), swung the 2016 election.
Wave 2: Service Sector Efficiency (2010-2022). Platform companies achieved enormous scale with minimal headcount. The gig economy replaced stable employment with on-demand work. Policy response: retrain service workers into knowledge work.
Wave 3: Cognitive-Work Displacement (2023-present). AI hits the occupations that were supposed to be safe — professional services, white-collar administration, creative industries, entry-level knowledge work. Anthropic's Observed Exposure study found 94% of workers have theoretical AI exposure. The most exposed? Older, educated, higher-paid. Not blue-collar workers. White-collar ones.
Each wave left an escape route. Factory workers could move to services. Service workers could learn to code. Knowledge workers could... what, exactly? Learn a trade?
The Escape Route That Isn't
That's exactly what 48% of executives are recommending. ResumeTemplates surveyed 933 executives in March 2026: nearly half said workers should "pursue trades over college." Sixty percent of Gen Z is now considering skilled trades.
The same executives are investing in humanoid robots. The Unitree R1 costs $4,900. Tesla has 1,000+ Optimus units deployed in its factories, targeting mass production. Figure AI's BotQ factory is building at a rate of 12,000 units per year.
The price of a humanoid robot fell 99.75% in three years — from $2 million to $4,900. The escape route from cognitive displacement leads directly into physical displacement. The waves are converging.
The BLS Can't See It
Perhaps the most damning implication of the Great Decoupling is how invisible it is to the measurement systems we rely on.
The BLS overcounted 2025 job creation by 3.2×. For an entire year, the Federal Reserve, Congress, and the public made decisions based on a labor market that didn't exist. The benchmark revision quietly explained the discrepancy — issues with the birth-death model, immigration estimation, survey response rates — but the damage was structural. Policy was made on fiction.
And the measurement failures compound. The Klarna Blueprint (the company eliminated 3,104 positions with zero WARN triggers and zero unemployment claims, via hiring freeze and attrition). Shadow agents deployed by employees without management approval. Pre-emptive displacement — companies cutting based on AI potential, not performance. None of these show up in the monthly jobs report. The Great Decoupling is happening faster than the instruments designed to measure it can detect.
The K-Shaped Economy
Mohamed El-Erian calls it the K-shaped economy. The letter tells the story: two arms diverging from a single point.
The rising arm: Fortune 500 profits at $1.87 trillion. S&P 500 margins expanding. Senior workers in AI-exposed fields seeing +8.5% wage growth. AI investment at $427 billion in Big Tech capex alone.
The falling arm: 181,000 total jobs created in 2025. Entry-level hiring down 13% in AI-exposed occupations, according to Brynjolfsson's "Canaries in the Coal Mine" data. Software developer entry-level positions collapsed from 43% to 28% of postings (Burning Glass). 52% of the Class of 2023 underemployed. U-6 underemployment at 8.7%.
GDP measures the sum of both arms. It looks fine. But the population on the falling arm is growing, and the population on the rising arm is shrinking.
The Dallas Fed Mechanism
The Dallas Fed provided the most rigorous explanation of how AI drives the K-shape. Their February 2026 analysis found that AI replaces codifiable knowledge — the kind you learn in school, the kind entry-level workers bring — while complementing tacit knowledge — the kind you accumulate through years of experience.
The numbers: employment in the top 10% of AI-exposed sectors has fallen 1% since ChatGPT's release, while overall employment grew 2.5%. But wages in those same sectors rose 8.5% — above the national average. How? The experienced workers who remain are more productive. The entry-level workers who would have joined never got hired.
This is the Great Decoupling at the individual level. GDP rises because the remaining workers produce more. Employment stagnates because the pipeline is blocked. Average wages go up because the composition shifts toward the more experienced. Every aggregate indicator looks "healthy" while an entire generation is locked out.
The computer systems design sector — ground zero for AI — tells it starkest: employment down 5% since November 2022, wages up. Fewer people, paid more, producing more. The K-shape made quantitative.
"This Time Is Different" — And It Might Actually Be True
Economists are trained to be skeptical of that phrase. And historically, technological disruption has always eventually created more jobs than it destroyed. Steam took a century. Electricity took fifty years. Computing took forty.
But AI is doing something none of those technologies did. It's improving its own design. GPT-4 to GPT-5.4 — the first model to exceed human performance at operating computers — took 24 months. The Delegation Flip — when enterprise AI usage crossed from majority-augmentation to majority-automation — happened in under nine months.
Previous technologies automated specific task types. AI automates across the entire spectrum — creative, analytical, administrative, communicative. Previous automation was visible. Factories closed. Machines replaced assembly lines. AI displacement is invisible: a hiring freeze, a contractor substitution, an employee who builds an agent that does the work of three colleagues who are never replaced.
And for the first time, cognitive and physical automation are converging simultaneously. $4,900 humanoids plus superhuman computer use equals no categorical refuge. The waves aren't sequential anymore. They're simultaneous.
The Social Contract Problem
The fundamental bargain of capitalism is that economic growth lifts all boats. GDP goes up, employment goes up, wages go up. That bargain held, roughly, from 1947 to 2000. The labor share was stable. Employment intensity was high enough that growth meant jobs.
That bargain is broken. GDP at 2.2%. Employment intensity at 0.05. Labor share at a 78-year low. The economy is growing. Workers are not participating in that growth. This isn't a policy failure. It's a structural transformation.
Goldman Sachs estimates AI automation could displace 25% of all work hours, with 6-7% of jobs eliminated and 1 million added to the unemployed at peak. Those numbers sound manageable until you realize they describe a permanent state, not a temporary adjustment. The jobs don't come back. GDP keeps growing anyway.
That's what permanent decoupling means. Not a crisis. A new normal.
The Bottom Line
The Great Decoupling is not a prediction. It is a description of the present, validated by 78 years of labor share data, 40 years of declining employment intensity, and one very quiet benchmark revision that rewrote the labor market history of 2025.
The economy can now grow without creating jobs. That single sentence breaks the assumption underneath almost every policy framework in existence — welfare-to-work, job training, even the Fed's dual mandate. All of them assume that a growing economy eventually means a growing labor market. The data says that assumption expired somewhere around 2001 and has been getting worse ever since.
A year from now, we'll get another benchmark revision. It will probably show that 2026 was weaker than reported, too. The instruments were built for a world where growth meant jobs. That world is over. The numbers just haven't caught up yet.