Two years after the burst of enthusiasm for generative artificial intelligence, the promises are still running faster than their results.
Millions of dollars flow into AI-branded models, data centers, and new products, while in economic reports, productivity scorecards, and corporate profits, the impact remains almost invisible.
“You see it everywhere, except in the statistics”
By: Gabriel E. Levy B.
In 1987, the economist Robert Solow issued a sentence that, due to its lucidity, is still valid almost four decades later: “You can see the age of computers everywhere, except in productivity statistics.”
This observation, as ironic as it is accurate, was made at a time when information technology promised to change the very structure of the global economy, but it still did not leave a mark on the key indicators of growth.
Analyst JP Gownder, vice president and principal analyst at Forrester, sums it up with a blunt reference: “Solow’s Paradox seems to be repeating itself.” And if history repeats itself, disenchantment will soon take center stage.
It took computing nearly two decades to overcome Solow’s Paradox. Although personal computers began to become popular in the 1980s, it wasn’t until the early 2000s that their impact on productivity became apparent.
The key was not only technological adoption, but the profound transformation of business processes.
Companies had to redesign workflows, train their employees, and adapt their business models to truly harness computational power.
When these conditions matured, information technology ceased to be just a support tool and became a real engine of economic growth.
The figures accompanied his diagnosis. According to data from the U.S. Bureau of Labor Statistics, between 1947 and 1973 (a pre-digital era), labor productivity grew by 2.7% annually. By contrast, between 1990 and 2001, when personal computers were already part of offices and homes, that figure barely reached 2.1%. Between 2007 and 2019, even with the internet fully integrated into everyday work, it fell to a modest 1.5%.
Solow’s paradox thus became a kind of ghost for each new technological revolution. And now, in the face of the rise of generative artificial intelligence, it is once again present.
A lot of noise, little profit
The current figures reinforce the suspicion that the impact of artificial intelligence is still not reflected in hard data.
Gownder puts it bluntly: “We’re just not seeing them.”
While tech giants invest billions in specialized chips, model training, and infrastructure expansion, productivity is not seeing a clear jump.
Although the third quarter of 2025 showed a 4.9% improvement in labor productivity in the non-agricultural business sector, this one-off rebound is not enough to sustain a narrative of structural transformation.
Even more worryingly, according to a recent study by the MIT Media Lab, 95% of generative AI business projects fail to generate a measurable impact in financial terms.
The investment, for now, seems to be more associated with a search for strategic positioning or avoiding being left behind than with a tangible return.
This phenomenon not only worries financial analysts, it also challenges technologists, governments and workers.
The returns that have not yet arrived
The overflowing enthusiasm with which companies received generative artificial intelligence during 2023 and 2024 turned into a kind of race to adopt tools, from conversational assistants to content automation systems, through predictive analytics and personalization of services.
In theory, these implementations should reduce costs, improve operational efficiency, and free up human capital for higher value-added tasks.
But, in practice, the implementation was much more complex.
The systems require specific training, constant supervision and adaptation to the particular environment of each company.
Many executives, seduced by spectacular demonstrations at conferences or viral articles, discovered that generative tools were not plug-and-play.
Integration time, misinterpretation, algorithmic biases, and a lack of alignment with internal processes slowed down the promised benefits.
At the same time, the expectation that AI would replace large-scale human tasks, a fear shared by millions of workers, did not quite materialize.
In many cases, people simply changed their workflow, but they didn’t disappear from the production process.
The work increased in complexity, not speed. The promises of efficiency were faced with a reality that was less malleable than the models seemed to anticipate.
In addition, a structural dilemma arises: who captures the value generated by AI? If the benefits are concentrated in a handful of companies capable of developing or deploying these systems at scale, the aggregate economic effect will be limited.
This also happened in the age of personal computers: while manufacturers and developers were winning, many traditional industries were barely transforming.
Corporate promises, failures and silences
The most illustrative cases of disenchantment with AI are repeated in different sectors.
In the legal industry, several firms have tried integrating automatic drafting systems for contracts and opinions.
However, semantic errors and the difficulty of contextualizing specific cases ended up generating more revision work than saving time.
In the medical field, diagnostic aid systems showed mixed results.
While some clinics achieved marginal improvements in diagnostic accuracy, others faced ethical and operational issues related to algorithmic decision-making.
In several cases, medical staff chose to deactivate the systems after noticing inconsistencies with established clinical criteria.
In the financial sector, banks and insurers invested in AI-powered risk analysis and fraud detection tools.
The results, in general, were positive in very specific tasks, but far from a comprehensive transformation of the business.
According to internal data leaked in 2024 by a Silicon Valley investment firm, only 8% of companies that implemented generative AI in customer service processes reported a significant improvement in user satisfaction.
And we must not forget the silences.
Numerous companies that in 2023 announced with great fanfare the mass adoption of AI have stopped mentioning their initiatives in their 2025 earnings reports.
The media enthusiasm was replaced by discreet prudence, a clear symptom that the returns did not justify the noise.
In conclusion, generative artificial intelligence has not yet fulfilled its promise of energizing the economy on a large scale. As with personal computers in the past, technology advances, but its real benefits take time to manifest. This does not imply that AI has no potential, but that its impact requires time, adjustments, and realism. Solow’s Paradox, more than a definitive judgment, is a call to look beyond enthusiasm and evaluate, with data in hand, when a technological revolution really becomes an economic transformation.
References
- Solow, R. M. (1987). We’d better watch out. New York Times Book Review, July 12.
- U.S. Bureau of Labor Statistics (2025). Labor Productivity and Costs, Third Quarter 2025. https://www.bls.gov/news.release/prod2.nr0.htm
- MIT Media Lab (2024). Generative AI in Business: A Performance Review. Massachusetts Institute of Technology. https://www.media.mit.edu
- The Register (2026). Forrester analyst: We’re still waiting for AI productivity gains. https://www.theregister.com
