AI and the productivity boost: the evidence meets the hype?

The theme of productivity has always been intertwined with that of work, giving the debate a bittersweet flavor. While an increase in productivity underpins economic growth, it can also imply a significant loss of jobs.

Precisely as a result of the rapid advances in AI, in an interview on CNN shortly before last Christmas, the founding father of deep learning and Nobel laureate in physics Geoffrey Hinton predicted that there could be massive job losses as early as 2026. Back in 2016, the British scientist had already stated that radiology should no longer be taught, because within a few years radiologists would be replaced by machines. That prediction has so far proven unfounded: not only have radiologists not disappeared, but in the meantime they have become the second-highest-paid medical specialty in the United States, and medical schools around the world are more crowded than ever with young people eager to join their ranks. And all this has happened not despite AI, but thanks to it: AI has made it possible to increase radiologists’ productivity more than that of other medical professions (which is why salaries have risen proportionally more).

In a recent post on X, viewed by millions of users, Aaron Levie—an entrepreneur and one of the most closely followed technology experts in Silicon Valley—spoke of a Jevons paradox for knowledge work, predicting a true explosion of it in the coming year. William Stanley Jevons, a nineteenth-century English economist and one of the forerunners of the marginalist school, was the first to observe, shortly after the mid-nineteenth century, that greater efficiency in the use of coal did not lead to a reduction in demand but paradoxically to an increase, as it made coal more attractive for a range of downstream applications that previously could not even be considered. A century and a half later, the paradox is well known in the environmental literature, unfortunately in a negative sense. Greater efficiency in energy use can in fact lead to a rebound in consumption, partially offsetting the benefits for the environment (and for energy bills). In the case of AI, however, the Jevons paradox would be the best and most immediate insurance against the catastrophism of those who imagine that, to be productive, technology must lead to mass layoffs (or at least to the consolidation of the lack of new hiring observed in the United States in several studies published over the past year). For them, the Holy Grail of productivity—long sought and repeatedly elusive for economic systems following the various information revolutions from the postwar period to today (with a single partial exception in the case of the internet around the turn of the century)—could only be achieved at the price of an inversely proportional change in employment. In reality, this does not necessarily have to be the case, and the Jevons paradox, coupled with a technology that is still imperfect after all, is there to prove it.

In Levie’s concrete example: “When you’re a small team, you’re making decisions between having a good marketing website, building a new product experience, handling customer support requests, taking care of something important on the finance side, finding new distribution channels, and so on. Each of these areas of investment and time competes with the others, and together they all end up slowing growth. Today, however, we have the opportunity to blow up the main constraint that drives many of these trade-offs: the cost of doing these activities… Demand will increase by 10 or 100 times in many areas of work, because we have lowered many of the other barriers to entry that prevented experimentation with many types of activities that most companies would never even have considered before… Naturally, many people ask what will happen to all the jobs in this new world. The reality is that, despite all the activities that AI allows us to automate, people are still needed to put together the entire workflow and produce real value. AI agents require management, supervision, and meaningful context to deliver their full benefits. All the performance improvements in AI models over recent years have led to higher-quality output, but we are still very far from a fully autonomous AI that perfectly implements and maintains what we are trying to achieve.”

From this perspective, therefore, productivity will increase, but not so much as to make human labor superfluous. As agentic AI finds its way into companies and, more generally, into various organizations, the role of employed people will increasingly shift toward the ability to assign tasks to agents and to verify their results—at least in functions sufficiently exposed to the impacts of technology (which already today constitute the majority).

According to some observers, the latest data on U.S. GDP for the third quarter of 2025, when compared with labor market trends, could confirm a tendency toward rising productivity that may finally be detectable at the macroeconomic level. Putting together, on the one hand, GDP growth at an annualized rate well above expectations, at 4.3%, with a (nearly) simultaneous slowdown in employment growth and a (slight) increase in unemployment—and with the necessary caveat that statistical measurements of macroeconomic variables are often imperfect and therefore subject to frequent ex post revisions—some observers advanced the hypothesis of an AI-boosted GDP. After all, to justify possible future rate cuts, the designated chair of the Federal Reserve, Kevin Warsh, has stated that thanks to AI the U.S. economy will be able to grow at faster-than-normal rates without generating inflation—another trade-off that is particularly central in economics. However, this is not only a U.S. debate.

In a recent study by the International Monetary Fund, Florian Misch, Ben Park, Carlo Pizzinelli, and Galen Sher asked whether artificial intelligence can provide a boost to Europe’s stagnating productivity, noting that achieving large benefits will depend on European countries’ commitment to growth-friendly reforms and on their willingness to keep regulation flexible, so as to help the new technology thrive. In the absence of reforms, their research shows that the medium-term productivity gain from AI alone would vary considerably across countries and, for Europe as a whole, would be rather modest: about 1.1 percent cumulatively over five years. With pro-growth reforms, however, much larger gains are possible in the long term, amounting to roughly double that figure—a signal of how context matter for truly AI-driven growth.