Stanford Economist Says AI Productivity Surge Is Finally Here

Erik Brynjolfsson, director of Stanford's Digital Economy Lab and one of the original voices studying AI economics, published a Financial Times op-ed on February 15th declaring "The AI productivity take-off is finally visible." His evidence? U.S. productivity jumped roughly 2.7% in 2025—nearly double the 1.4% annual average from the past decade.

There's just one problem with celebrating this productivity "harvest phase": the Bureau of Labor Statistics revised 2025 job gains down to just 181,000, from an initial print of 584,000. That's a downward revision of 403,000 jobs. The economy grew at 3.7% in Q4 while adding almost no workers. Which means either we've achieved remarkable efficiency gains, or we're measuring the wrong things.

Probably both. And the distinction matters more than Brynjolfsson's optimistic framing suggests.

The J-Curve Theory: Investment Pain, Then Productivity Gain

The J-curve concept holds that general-purpose technologies like AI don't produce immediate benefits. Massive investment comes first—in infrastructure, training, reorganization—obscuring early gains. Productivity initially appears to decline as companies spend heavily without visible returns. Only after this investment phase does productivity take off, creating the J shape.

Brynjolfsson argues we're transitioning from the investment phase to the "harvest phase where those earlier efforts begin to manifest as measurable output." The 2.7% productivity jump in 2025 represents this shift becoming visible in macroeconomic data.

Not everyone agrees that the transformation is happening. Apollo Chief Economist Torsten Slok quipped that "AI is everywhere except in the incoming macroeconomic data," echoing Robert Solow's famous observation about the PC revolution: "You can see the computer age everywhere but in the productivity statistics."

Slok noted that employment, inflation, profit margins, and earnings forecasts for S&P 500 companies outside the "Magnificent 7" still lack clear evidence of AI's impact. "Maybe there is a J-curve effect for AI, where it takes time for AI to show up in the macro data. Maybe not," he wrote.

The question isn't whether AI is being deployed—it obviously is. The question is whether deployment is generating the productivity gains that justify the investment, or whether we're measuring output increases while missing the human costs.

What Productivity Numbers Actually Measure

Productivity is output per worker. When GDP grows 3.7% while employment increases just 181,000—down from 1.46 million in 2024—productivity calculations show impressive gains. But this measures efficiency, not value creation or societal benefit.

Consider what the numbers don't capture:

Job quality: If AI eliminates 400,000 mid-skill positions while creating 181,000 new jobs requiring different skills, the productivity number looks great. The workers whose jobs disappeared aren't included in the calculation.

Income distribution: Productivity gains flowing entirely to capital owners and high-skill workers produce different economic outcomes than gains shared broadly. The 2.7% number doesn't distinguish.

Sustainability: If companies achieve short-term efficiency by cutting workers faster than AI can genuinely replace their output, initial productivity spikes may not persist. We're measuring the beginning of a transition, not the end state.

Transition costs: Workers displaced by AI face retraining costs, potential wage cuts in new roles, geographic relocation, and career disruption. These show up as personal costs, not in aggregate productivity statistics.

Brynjolfsson acknowledges these limitations, noting that "several more periods of sustained growth are needed to confirm a long-term trend in productivity." But the celebration of 2.7% gains without equal attention to the 403,000 job revisions tells you where the emphasis lies.

The Power Users Brynjolfsson Found

Brynjolfsson's research reveals "a small cohort of power users" automating end-to-end workstreams with AI agents, completing tasks in hours instead of weeks. This matters because it suggests productivity gains aren't evenly distributed—they concentrate among organizations sophisticated enough to implement AI comprehensively.

His 2025 study showed AI hitting entry-level workers disproportionately, especially those ages 22-25 in highly AI-exposed professions. That's the demographic traditionally using entry-level positions to build skills, professional networks, and career trajectories. If AI eliminates those positions before workers develop higher-level capabilities, you don't get smooth transitions to new roles—you get permanent displacement.

The "power users" automating workflows represent one end of the distribution. The 181,000 net new jobs represent the other end of the spectrum. In between is the vast middle of companies "still using AI in minimal ways"—experimenting with tools, achieving incremental gains, not yet seeing the transformative productivity Brynjolfsson describes.

This distribution pattern matters for policy. If productivity gains concentrate among sophisticated early adopters while job losses distribute broadly, the aggregate numbers look great, while individual outcomes diverge dramatically. That's exactly what a K-shaped economic recovery describes—winners pulling away while others fall behind.

The ICT Industry as a Leading Indicator

Stephen Brown, chief deputy North America economist at Capital Economics, points to information and communication technology (ICT) industries as showing clear AI productivity signals. ICT output rose during Q3 2025 despite employment dropping.

Brown notes that while earlier payroll cuts were likely a pandemic overcorrection, reductions have continued even as ICT sectors boomed. "All this implies that AI is making a large contribution to productivity growth," he concluded.

This is the most straightforward evidence: sectors that directly deploy AI are producing more output with fewer workers. The productivity gains are real and measurable. The question is what happens to the workers whose output is no longer needed.

The ICT industry employs roughly 5.4 million workers in the U.S., according to BLS data. If AI enables continued output growth while employment declines, those workers either transition to other sectors (possibly at lower wages), retrain for higher-skill roles (if opportunities exist), or exit the workforce. The productivity statistics don't track which outcome dominates.

The Transition Brynjolfsson Isn't Emphasizing

"We are transitioning from an era of AI experimentation to one of structural utility," Brynjolfsson wrote. That's correct. Companies are moving from testing AI tools to implementing them as core business infrastructure. This generates the productivity gains he's measuring.

But "structural utility" refers to permanent changes in labor requirements. When workflows automate end-to-end, the jobs they replace don't come back. When entry-level positions disappear, career ladders break. When output increases while employment stagnates, the gains accrue to those who own the AI systems, not necessarily to workers.

Brynjolfsson's framing—"The productivity revival is not just an indicator of the power of AI. It is a wake-up call to focus on the coming economic transformation,"—treats this as an opportunity requiring attention. That's technically accurate but misses the urgency. The transformation is already happening. The 403,000 job revision isn't a future concern—it's a current reality.

What 2.7% Productivity Growth Actually Means

Doubling productivity growth from 1.4% to 2.7% annually sounds transformative. Over a decade, 1.4% compounds to 15% total growth. At 2.7%, it's 31% growth. That's the difference between incremental improvement and genuine economic transformation.

But productivity growth doesn't automatically translate to broad prosperity. The U.S. experienced strong productivity growth from 1995 to 2005 during the Internet boom. Median wages barely moved during the same period. The gains concentrated among high-skill workers and capital owners, while median workers saw minimal benefit.

According to data from the Economic Policy Institute, productivity grew by 72% between 1973 and 2014, while median hourly compensation grew by just 9%. The relationship between productivity and worker outcomes broke decades ago. There's no reason to assume AI-driven productivity gains will be distributed differently without policy interventions that force broader sharing.

Brynjolfsson's 2.7% productivity figure suggests AI is working from an efficiency perspective. It doesn't tell us whether that efficiency produces broadly shared prosperity or concentrated gains with widespread displacement.

The Geopolitical And Monetary Risks Brynjolfsson Mentions

Brynjolfsson cautioned that "geopolitical or monetary snafus could offset advances." This deserves more attention than a throwaway line. AI productivity gains require stable economic conditions, functioning global supply chains, and policy frameworks that support both innovation and transition management.

Geopolitical tensions around semiconductor supply, AI regulation divergence between the U.S., the EU, and China, and trade policies affecting technology transfer all create risks to sustained AI deployment. Monetary policy mistakes—either maintaining rates too high and choking investment, or cutting too aggressively and reigniting inflation—could disrupt the capital availability needed to fund AI infrastructure.

These aren't hypothetical risks. They're active threats to the "harvest phase" Brynjolfsson describes. The productivity gains are real but fragile, dependent on continued investment that requires stable economic and geopolitical conditions.

What This Actually Tells Us

The evidence increasingly supports the claim that AI is generating measurable productivity gains. The 2.7% productivity growth in 2025, the ICT sector's output increase despite employment declines, and Brynjolfsson's research on power users all point in the same direction: AI is working as an efficiency technology.

The question is: efficiency for whom? The 181,000 net new jobs in 2025—down from 1.46 million in 2024—suggest efficiency gains are producing output growth without corresponding employment growth. That's productivity by definition. It's also displacement by consequence.

Brynjolfsson's optimistic framing—transitioning from investment to harvest phase, productivity revival as wake-up call—treats this as an opportunity requiring focus. The workers whose jobs disappeared in the 403,000 downward revision probably see it differently.

Both perspectives are valid. AI is generating productivity gains that could support higher living standards and economic growth. AI is also concentrating those gains while displacing workers who lack clear transition paths to comparable roles. These aren't contradictory—they're simultaneous realities of the same transformation.

The policy challenge is to ensure that productivity gains translate into broadly shared prosperity rather than concentrated wealth alongside widespread displacement. Celebrating 2.7% productivity growth without addressing the 403,000 job revision treats half the equation as the whole story.

Understanding AI's economic impact requires measuring both efficiency gains and transition costs—not just aggregate productivity numbers. Winsome Marketing's growth experts help you evaluate AI implementation through the lens of actual workforce impact, not just output optimization. Let's talk about AI strategies that balance productivity with sustainability.