The Unproven Promise of AI
The Unproven Promise of AI
Highlights
AI’s impact on economic output is debated, and likely minimal to date
The share of economic output given to workers is the lowest level on record
56.7% of employers report having an AI policy, yet only 19% of workers report receiving extensive training, leaving nearly half to learn on their own
Workers report they are working harder, learning more, and expecting more, all while the share of output flowing to paychecks shrinks
According to the Federal Reserve’s February Beige Book, businesses are integrating AI to augment output rather than replace jobs. It is tempting to link recent GDP growth, which has diverged from a softening labor market, to productivity gains from AI. Yet despite sensational headlines, AI’s real-world business impact remains limited. And the most important question may not be how AI affects productivity, but who benefits if and when it does.
An MIT study of hundreds of businesses found that despite tens of millions of dollars in GenAI investments, 95% of companies saw no financial return. These tools can help individual workers perform tasks more efficiently, but that rarely translates into measurable gains at the business level. There is a meaningful difference between employees using Gemini or Claude at their desks and companies unlocking the kind of systematic efficiencies that move the needle on firm performance. As the MIT study puts it, “adoption is high, but disruption is low.”
New research from Goldman Sachs tells a similar story of AI having a minimal impact. AI infrastructure investments, which some estimates point to as a significant driver of growth in early 2025 due to large-scale high-cost builds, had virtually no influence on GDP gains. While experts debate AI’s true impact, both on productivity and on infrastructure investments, the majority of recent GDP growth is more likely to be driven by an increase in consumer spending.
Workers’ slice of the economic pie is shrinking
What is not up for debate is how productivity gains make their way, or rather don't make their way, into workers' pockets. The labor share of output, the percentage of economic growth that ends up in paychecks rather than corporate profits, has been falling steeply since the end of the dot-com boom and now sits at its lowest level on record. Technological advancements have increased output for decades, but that output is increasingly decoupled from wages as businesses direct more gains toward capital. Automation has consistently followed this pattern: it drives capital investment and productivity higher, and as a result, a smaller share of output flows toward wages.
Some interpret current trends as evidence that AI is already displacing workers and accelerating the K-shaped economy. However, a closer look at market data provides a different explanation: the falling labor share is better captured by identifying who is benefiting from the gains rather than by a reduction in total employment. Economic indices confirm that the labor share declines when wages grow more slowly than profits, not necessarily due to fewer people working. For instance, the S&P 500 is up 15.5% year-over-year from last February, while BLS data reveals that wages have risen just 3.8% in the same period, illustrating where the gains aren’t flowing. Although factors like an aging population and reduced immigration make headline employment figures harder to interpret cleanly, employment data for prime-age workers remains stable, and layoffs remain low. These indicators suggest that worker displacement is not the primary force behind the decline in labor share.
The more likely culprit is the same one that has defined every technology cycle since industrialization: automation expands the pie, but that doesn’t necessarily mean workers end up with an equally larger slice. While there are cases where new technology, including AI, could boost demand for high-skilled work and even lead to higher wages in specific sectors, these examples remain the exception rather than the rule. AI’s real threat to workers may be less about displacement and more about speed—a shift that moves faster than wages, training, or policy can keep up with.
Workers don't need an economics textbook to feel this. Despite strong gains in economic output, the average worker remains far removed from GDP growth. According to ZipRecruiter survey data on worker sentiment, just under a third of workers report their pay has not kept pace with inflation over the last two years, and over half report pay increases of 5% or less in that period. The consequences extend well beyond paychecks: 34.3% of workers aged 18 to 34 have delayed buying a home in the past year, and 21.1% have delayed starting a family due to rising costs. This growing gap between economic gains and wage growth is directly reflected in these postponed life milestones, showing how a shrinking labor share is felt not just at work but in nearly every major decision facing young workers. Economic growth is strong, but workers are struggling.
Expectations on a collision course with reality
That gap between growth and compensation doesn't close on its own. Employers are framing AI as a workforce opportunity to increase output, but that opportunity is not fully realized, and its benefits are not evenly distributed. The institutional support needed to unlock the promise of AI is largely absent, at least in practice. Employer survey data tells one story: 56.7% report having an internal AI policy, and 54.4% actively encourage AI use among employees. Worker survey data tells another: only 19% report receiving extensive AI training from their employer, and nearly half (47.5%) say they are learning these skills entirely on their own time. The intent exists, but the implementation is still catching up.
This creates a hidden burden for workers, who absorb the transition costs. Workers are investing real effort in exchange for expected returns: 68.2% anticipate a promotion within the next three years, and nearly half (47.9%) feel their compensation doesn't reflect the work they actually do, whether they're being paid below market rate or quietly performing a higher-level role without the title or pay to match. That optimism isn't unfounded, given that 85.7% describe themselves as engaged or overachieving, but it is increasingly disconnected from where the gains from that effort are actually going. The structural trend is working against them. Workers report they are working harder, learning more, and expecting more, all while the share of output flowing to paychecks continues to shrink. That is a defining tension of the AI era: not job loss, but the quiet erosion of what work is actually worth.
Technology has never been the obstacle to shared prosperity. Distribution has. AI will not change that dynamic on its own. Without deliberate investment in workers, workplace infrastructure, and AI policy, the next leap in productivity will follow the same path as every one before it: gains that flow to capital, and a bill that gets handed to labor.
To chart a different course, policymakers and business leaders can take concrete steps now. For instance, targeted tax incentives or grant programs for companies that provide substantive AI skills training to their workforce could help ensure that the benefits of new technologies flow more broadly. Mandating measurable outcomes for worker upskilling initiatives and supporting labor-management partnerships, which put workers at the center of internal AI policy, can further align growth with widely shared progress. Turning insight into action means making deliberate choices to invest in humans, not just capital.
Methodology
Survey data in this research are sourced from the following:
ZipRecruiter surveyed 1,500 employed individuals across U.S. businesses of various sizes and industries from January 5-14, 2026, to learn how AI is influencing their jobs.
ZipRecruiter surveyed 1,500 employers across the U.S. from September 8-16, 2025, to gain insight into hiring practices and AI integration within businesses.