Feeling a little disappointed by AI’s transformation of the economy? The so-called “fifth industrial revolution” is set to eliminate half or even all of the workforce, and yet for many workers its adoption is reluctant, even optional. For those who have started doing it, it often feels like homework – it writes your emails for you, but often it’s still wrong. Three and four decades ago, the computer revolution was similarly hyped, and yet for a long time it looked more like Pets.com than what later became the iPhone.
That’s the view taken by Goldman Sachs’ Elsie Peng, who delved into the rise in productivity brought on by the computer revolution in a research note for the bank earlier this month. The last time something this big happened, she wrote, things actually got measurably worse before they got better — for four years, by her estimate. Then they flatlined for another four. It wasn’t until the eighth year that the gains through math became statistically significant. The productivity boom that everyone associates with the personal computer did not show up in macro data until 15 years after the commercialization of the PC.
Goldman’s official view is still that AI will “significantly boost productivity growth over the next decade.” What Peng is arguing is that the drivers of the AI boom – and many investors who are pricing that boom into stocks – may be completely mistimed. And the reason for this, in their opinion, lies less in the technology and more in the people who are supposed to use it.
Table of Contents
Nobody mentions the J-curve
The personal computer was commercialized in 1981. In the early 1980s, investments in information and communications technology increased sharply in most industries. Nevertheless, the productivity trend was flat until the late 1990s.
Peng’s industry panel analysis finds that the impact on productivity followed what she calls a J-curve: a moderate decline in the first four years, statistically significant gains only after eight years, and a peak of about 0.6 percentage points in year 12. If the launch of ChatGPT in 2022 corresponds to the debut of the personal computer in 1981, the J-curve assumes that the productivity payoff will arrive around 2030 at the earliest and will peak in 2034.
The first time around, three forces caused the deficit. Key components such as semiconductors and telecommunications equipment remained expensive throughout the 1980s and only declined when previously concentrated markets were opened up through regulatory intervention and increased competition in the 1990s. Many applications—particularly the Internet—only generated value after adoption reached critical mass, which took years. But the biggest bottleneck was something less visible: the enormous reorganization effort required to actually use the technology.
Goldman estimates that every dollar of investment in ICT hardware required at least $1.70 in complementary “intangible” investments – software, data systems and the hardest category to measure, organizational overhaul. Crucially, restructuring spending only increased in the mid-1990s, a full decade after personal computers hit desks. The industries that ultimately benefited most from ICT were not those that adopted ICT earliest or invested the most in hardware, but rather those that invested the most in redesigning work processes.
The divide is repeating itself – and the people problem is even worse
This is where the historical parallel becomes apparent. Goldman data shows that investment in AI hardware is already growing faster than ICT expansion in the corresponding phase. That’s the good news for the bulls. The bad news: investments in reorganizing work processes appear to be progressing more slowly than in the ICT cycle of the 1990s. Goldman acknowledges that some of this spending may not be captured in official statistics – an Atlanta Fed survey projects around $280 billion in AI-related intangible spending in 2026 – but even accounting for measurement gaps, the reorganization side of the general ledger is significantly behind the hardware side than it was last time.
The workforce appears to be voting with their feet on this restructuring that could one day lead to the promised land of AI productivity: They are fighting back.
A survey of 2,400 knowledge workers conducted in April by AI firms Writer and Workplace Intelligence—both firms with a commercial interest in AI adoption, so take the numbers with appropriate skepticism—found that 29% of employees admit to actively sabotaging their company’s AI strategy. Among Gen Z workers, that figure was 44%, up from 41% last year. A separate WalkMe survey of executives and employees in 14 countries conducted the same month found that more than 54% of workers had bypassed their company’s AI tools in the last 30 days to do the work manually instead. The survey commissioners found that much of this active and passive and even passive-aggressive resistance is driven by the “fear of obsolescence.” Similar, The Economist reported that after an early surge, AI usage among U.S. workers actually declined as initial enthusiasm waned.
Researchers at Harvard Business School have documented a phenomenon they call “symbolic adoption”: Instead of openly rejecting AI tools—which carries the risk of being fired—employees obey superficially while quietly undermining the technology behind the scenes. The motive is not mysterious. Of the AI saboteurs self-described in the Writer survey, 30% say they don’t want AI to take their jobs; 26% say technology has reduced their sense of value or creativity at work. These aren’t irrational answers, considering that 69% of executives in the same survey say their companies are already conducting AI-related layoffs.
What the data already shows
In my own reporting on AI’s impact on the job market, I’ve seen this dynamic reflected in the numbers. Erik Brynjolfsson of Stanford and ADP Research have begun tracking 4.6 million workers in more than 730 occupations through the Canaries Dashboard. What they find is not a calm overall labor market. Employment of workers ages 22 to 25 in AI-exposed occupations is shrinking by more than 4% annually, which is invisible at the headline level and only becomes visible when cutting by age and task exposure.
Goldman’s industry rankings suggest that information, professional services, insurance and finance are best positioned for early productivity gains – the same sectors of the economy where the sabotage surveys find the highest rates of resistance. If the reorganization lag is exacerbated by organized friction and not just cost and complexity, Goldman’s 8-12 year timeline derived from the ICT era could prove optimistic.
This pattern is consistent with the J-curve that Peng identified. In the early stages, technology changes faster than it creates. Productivity gains are being suppressed not because the technology doesn’t work, but because people haven’t reorganized around them – and in this cycle a significant proportion of them are actively resisting.
The Economist I didn’t report that AI usage has declined because the tools have gotten worse. It has fallen because adoption is difficult and adoption of technologies that workers associate with their own displacement is even more difficult. So far, experts in neuroscience and AI believe that the difficulty of old dogs learning new tricks is significantly underestimated. Joshua Wöhle, the CEO of Mindstone, a company that provides AI upskilling and reskilling services, said previously Assets In his experience, “most people hate learning. They would avoid it if they could.” Neuroscientist Vivienne Ming also spoke Assets about how she sees a difference between “problems well posed and problems badly posed,” and that unfortunately most of the training and work is focused on the former.
This particularly ill-posed problem could be a very big problem for financial markets, argued Torsten Slok, chief economist at Apollo Global Management, since AI was “the only thing that kept both the economy and the markets going.” With so much money in so few names, he wrote in his Daily Spark blog earlier this month: “A slower payout would not only be a sector problem, it would also risk tipping the economy into recession and the S&P 500 into a correction.”
What Goldman actually says
Of course, Goldman isn’t saying that AI is a mirage. The micro-level productivity gains from AI in specific applications are well documented. The question is when and if they will show up in the macro data – the kind of broad-based productivity acceleration that would justify current stock valuations and the level of infrastructure investment underway.
The historical record says: later than you think. The human resistance data says: probably later than Goldman’s own model assumed. And the reorganization gap says: The bottleneck is not the hardware, it never was. It is the messy, expensive and politically sensitive work of getting organizations and the people within them to actually change the way they work.
It took a decade last time. There is no obvious reason to believe it is happening any faster, and there is evidence that it may be happening more slowly.
https://fortune.com/2026/07/14/goldman-ai-productivity-reality-check-worker-sabotage/
