The AI productivity boom is not here (yet)
Artificial intelligence is improving fast. Its effect on output, not so much
Artificial intelligence is advancing at a breathtaking pace. The latest models can now complete complex, hours-long tasks with minimal human oversight. This month, one of OpenAI’s models even helped derive a new result in theoretical physics. It’s no wonder an essay proclaiming that “Something Big is Happening” has gone viral.
But is something equally big happening in the economy? Last year, Scott Bessent, the United States’ treasury secretary, predicted that AI would soon start “biting,” meaning it would deliver noticeable productivity gains. Kevin Warsh, President Donald Trump’s nominee to lead the Federal Reserve, is counting on an AI-driven productivity surge to help tame inflation.
At first glance, recent U.S. macroeconomic data seem to support their optimism. The economy grew a solid 2.2% in 2025, according to figures released on February 20th, yet hiring slowed sharply, with employers adding only around 15,000 jobs per month—equivalent to annual employment growth of just 0.1%. This combination implies that each worker is producing more output.
Closer inspection, however, offers little evidence of substantial AI-fueled productivity gains. Real GDP grew at an annualized rate of just 1.4% in the fourth quarter of 2025, partly affected by a government shutdown. The recent gap between output and employment growth is not unusual: since 1950, the difference between the two has exceeded two percentage points in nearly one-third of years. Based on preliminary estimates of GDP growth and aggregate hours worked, productivity growth in 2025 was about 1.9%—just below the long-run average of roughly 2% and far short of the gains seen during the internet boom of the 1990s and 2000s.
Much of the recent GDP growth reflects a surge in investment, particularly in AI-related infrastructure. Jason Furman of Harvard University estimates that around 90% of GDP growth in the first half of 2025 came from spending on data centers and related capital projects. Adjusting for investment-driven output tells a similar story: research from the Federal Reserve Bank of San Francisco finds that underlying productivity gains, once investment effects are excluded, are close to zero. Labor-market dynamics reinforce this view. Tighter immigration policies have reduced labor-force growth, boosting average productivity by excluding many workers in relatively low-productivity sectors like farming and construction. A sharp decline in temporary employment has had a similar effect.
Economists would need to examine three factors to determine whether AI is truly boosting productivity: how widely it is adopted, how intensively it is used, and how much it improves output when applied to specific tasks.
Adoption is rising. A tracker by Alex Bick of the Federal Reserve Bank of St. Louis and colleagues found that 41% of American workers used generative AI at work in November 2025, up from 31% a year earlier. Other surveys reach similar conclusions. Jon Hartley of Stanford University and colleagues estimate usage rates increased from roughly 30% at the end of 2024 to 36% by the end of 2025.
But adoption alone does not guarantee productivity gains. How intensively AI is deployed matters. Mr. Bick found that only 13% of working-age adults used it daily, and the share of total work hours involving generative AI remains small, rising from 4.1% in late 2024 to 5.7% by mid-2025. Most usage is task-specific rather than fully automating processes. OpenAI’s models are primarily used for writing assistance and information queries, while Anthropic’s Claude mainly helps with coding tasks.
When AI is used, the benefits can be substantial. In 2023, Shakked Noy and Whitney Zhang of MIT found that ChatGPT reduced completion times for writing tasks by nearly 40%. A study of consultants at the Boston Consulting Group by Fabrizio Dell’Acqua of Harvard Business School and colleagues reported AI-driven productivity gains of 12–25% on realistic professional tasks. A broader review by Maria del Rio-Chanona of University College London found average productivity improvements of 15–30% in real-world settings.
Taking adoption, intensity, and efficiency together, a rough calculation suggests AI has so far had only a modest impact on overall productivity. Combining the growth in hours spent using generative AI with efficiency gains yields an estimated boost of just 0.25–0.5 percentage points to productivity growth over the past year. This is likely an optimistic estimate, assuming all time saved is used productively and that workers do not offset gains by shirking or producing lower-value output. Early evidence indicates a more mixed picture: some studies find that workers actually spend more time working when using AI, while others note that the technology sometimes produces low-quality output requiring correction.
This highlights a deeper limitation of claims that AI is driving a productivity boom. Historically, sustained gains occur not simply when workers adopt a new tool, but when firms reorganize production around it. Early factories became only slightly more efficient when steam engines were replaced with electric motors; the true revolution came decades later, after layouts and processes were redesigned. Similarly, productivity growth lagged for years after personal computers became widespread, accelerating only once firms leveraged the technology fully—particularly in retail, where computers transformed logistics and inventory management.
There is little sign that AI has reached this stage. A recent study by Ivan Yotzov of the Bank of England and co-authors found that executives spend just 1.5 hours a week using AI, and nine out of ten senior managers report no measurable improvement in labor productivity. In other words, organizational transformation has barely begun. While something big may indeed be happening with AI itself, its effects remain largely invisible in macroeconomic data—for now.