01 / 05
AI, Tractors, and the Slow Diffusion of Labor-Saving Devices

Blog Post | Science & Technology

AI, Tractors, and the Slow Diffusion of Labor-Saving Devices

Productivity and economic growth are a tide that lifts all boats, whether we're talking about agricultural machinery or ChatGPT.

Summary: The proliferation of large language models, such as ChatGPT, raises concerns about job displacement, particularly in professions like writing and law. However, historical transitions, like the shift from horses to tractors, suggest that the adoption of new technologies is gradual, with human labor and machines coexisting and evolving together. While AI may revolutionize certain sectors, its widespread impact on employment is likely to be evolutionary rather than revolutionary.


“You can see the computer age everywhere but in the productivity statistics,” wrote economist and Nobel laureate Robert Solow in a 1987 book review. While computing was making massive strides in the 1980s and ’90s—cumulating in the internet mania and dot-com bubble and its burst—it took a long time before we saw large, economy-wide impacts of this revolutionary general-purpose technology. Computers, internet, and (smart) screens have made the modern world unrecognizable from what it was just a few decades ago, but the process took a lot longer than most techno-optimists suggested then—and their followers are suggesting now.

Last year had a strong 1990s feel to it: 2023 became the year of AI worries, with ChatGPT and other large language models making a bid for disrupting all manner of industries. Writers, musicians, lawyers, editors, physicians, graphic designers, and more were some of the professions seemingly at risk of losing their livelihood to machines that quickly replicate their work at a fraction of the cost.

We should draw two conclusions from these large language models and past technological transitions: First, some of their features are truly remarkable and can very well compete with (some) lawyers, writers, musicians, or graphic designers over the next few years. Second, radical, labor-saving technologies are slow to diffuse across the economy, so these mainly middle-class professions are probably safe from mechanization and automation a while longer.

Take the iconic horses-to-tractor transition in agriculture over the 20th century. From invention and commercial availability, tractors took decades to overtake horses. The same was true for steam engines, that iconic energy invention powering the Industrial Revolution—which itself was both one of most radical and life-changing events in all of human history and pretty dull on account of its slow changes.

“Over the sweep of history the tractor has indeed had an immense impact on people’s lives,” The Economist said in December 2023, “but it conquered the world with a whimper, not a bang.” In all ages, tech innovations have defused across societies slowly: the old technology remains relevant for decades. Even in the 1930s, two to three decades after the first tractors were built, there was more equine than mechanical horsepower on American farms. Scholars usually put the cutoff point for when tractors overtook horses on farms after World War II, a generation or more after they were first employed.

The story in The Economist also highlights the connection to real wages and overall economy. Productivity and economic growth are a tide that lifts all boats. A rule of thumb is that labor-saving technology gets adopted only once it makes sense to replace labor with machines. That happens when the machines get good enough, the cost of running them falls sufficiently, or the alternative uses for labor get high enough.

For the first few decades of the tractor’s existence, they weren’t obviously better than the source of literal horsepower that preceded them. They were expensive to run, were not that powerful, and couldn’t do many things that a horse could. For one, they compelled farmers to purchase gasoline—a resource they didn’t have—rather than putting horses out to grazing on land or having children (or cheap labor) tend to the animals—resources that they did have. And during the Great Depression in America, cheap labor wasn’t that hard to come by.

In time, those trends reversed; real incomes and better-paid manufacturing jobs bid labor away from the farms, and the demographic transition to smaller families meant that children couldn’t be routinely relied upon to, for example, care for the animals. In economic speech, the optimal size of farms grew, and on these larger farms the now much better tractors came into their own right.

In contrast, the labor that large language models like ChatGPT purport to displace is often highly remunerated, while the costs of running the operations are pretty low, which suggest that artificial intelligence (AI) might displace some such workers.

Yann LeCun, chief AI scientist at Meta/Facebook, is less convinced. In an interview with Wired magazine’s Steven Levy, LeCun says that “chatbots can produce very fluent text with very good style. But they’re boring, and what they come up with can be completely false.” For Reason magazine last year, I reflected on the promise of ChatGPT along similar lines: “Bad writers, cheating students, lazy professors, and journalists echoing press releases have clearly met their match. But the more human, creative, authentic storytelling that is the center of all meaningful writing has not.”

The tractor attempted to compete directly with a farming industry routinely employing over a third of the labor force in the 1800s and early 1900s in the United States, constituting some 15 percent of gross domestic product. What is the total addressable market for large language models and by extension general AI? Statista estimates that the market size of AI in the United States, a $27 trillion economy, is some $100 billion—a fraction of a percent. Even if that grows seven-fold this decade in line with projections (nothing to scoff at!), that’s also not as revolutionary as most pundits have claimed in this the first year of their widespread adoption.

Solow, who died late in 2023, might have repeated his statement about this new form of the computer age. Like the computer in the 1990s, the onset of AI and large language models make more media puff and noise than they have real-world, real-economy impact.

There is some suggestion that adoption of new technology and consumer items happen somewhat faster today than, say, in the 1940s or 1980s, but not by a lot. Still, what is more likely to happen than mass white-collar unemployment is that we’re going to have a symbiosis for a while. The old human-labor technology and the new machines will coexist, and human workers will gradually employ the machines in different ways. An evolution, not a revolution.

At the end of the day, writes Mike Munger for the American Institute for Economic Research, ChatGPT is just a calculator. And like calculators did, it will obsolete some human labor. And that will be fine.

European Central Bank | Labor Productivity

AI Use Reaches a Majority of Workers in Europe

“Let’s look at AI’s footprint in Europe’s workplaces. There is an unmistakable trend. In just two years, the share of workers using AI on the job has doubled from 26% of survey respondents in 2024 to 41% in 2025, reaching 52% in 2026. A majority of workers now report using AI for work and, on average, they use it around three days per week.

the median user reports saving three hours per week, or about 7.7% of median working time, although this overall figure contains a highly skewed distribution. Most users enjoy moderate time-savings, while a small number report very high efficiency gains…

The time saved already suggests that AI makes workers more productive. But for the three hours saved – or 7.7% saved working time – context matters, if we are to accurately assess the implications for the total economy productivity growth.

First, only half of workers reported using and saving time thanks to AI (48.8%). Thus, for the whole economy, the overall efficiency gain (in other words, the share of savings in working hours attributed to the use of AI) is closer to 3.8%.

Second, it’s worth noting that these time-savings only translate into higher productivity if workers turn the freed hours into extra output, rather than using them for less directly productive activities. Crucially, the productivity boost also depends on whether the employer is in a position to put that extra capacity to use.”

From European Central Bank.

The Keyword | Employment

AI Is Helping Workers Across Jobs, but Not Replacing Them

“Google is launching the first iteration of the AI & Economy ATLAS (Activity, Task, Landscape, and Adoption Study), an ongoing, large-scale, de-identified study of how people are using Google’s AI products and tools…

Here a few of the most interesting observations so far:

  • AI use at work is broad but shallow: Workplace adoption spans all industry sectors and also 68% of all occupations that collectively represent 90% of total U.S. employment. However within jobs, people are using AI selectively: in a typical job AI is used for only ~21% of tasks.
  • At work, most AI use is focused on collaboration and assistance with tasks, and so far task automation is uncommon: ATLAS data shows the vast majority of AI interactions at work focus on collaborative uses such as ideation, strategy, information retrieval, and learning. Tasks like creative design and hypothesis testing (categorized in ATLAS as “non-routine cognitive”) show up in AI work interactions at a much higher rate than in the economy as a whole (65% vs 35%). Less than 10% of those interactions fully automate tasks.
  • AI use is not limited to white collar workers, it’s also assisting workers in predominantly physical and manual occupations with adjacent tasks: AI use for work is not limited to jobs traditionally seen as knowledge work. While not as prevalent, workers in manual and technical trades (e.g., auto technicians, industrial mechanics) are using conversational AI as a live collaborator for real-time diagnostics, troubleshooting, and on-the-fly learning. When workers in these areas use our AI tools, they’re 2x more likely to use multimodal AI (i.e. using AI to create images or video). For example, automotive technicians and industrial mechanics use AI to interpret complex test results, debug electrical wiring, and inspect machinery for wear.
  • AI is delivering value at home that may be missed in standard economic metrics, particularly around high-friction administrative tasks: Over 86% of interactions with AI tools in ATLAS occur outside of work. People are using AI in new and interesting ways not captured in standard economic metrics including productive household activities (e.g. researching purchases, help with using appliances, and tools) and high-friction administrative tasks (e.g. navigating government services like taxes, licensing, and fines).”

From The Keyword.

Blog Post | Employment

Evolving Markets Drove the Remote Work Revolution

Free choices—not mandates—have made flexible work a lasting reality.

Summary: Remote and hybrid work have become a durable feature of the modern workplace, with most remote-capable employees choosing flexible arrangements that benefit both workers and employers. Evidence suggests these models can improve productivity, job satisfaction, health, and employee retention while reflecting voluntary market decisions rather than government mandates. As with many past workplace improvements, adoption has largely preceded legal recognition, suggesting that policy should remove barriers to flexible work instead of attempting to direct it.


Back in 2020, I noted, “The dramatic rise in telework amid the pandemic is a radical experiment, but its effects will be long-lasting.” I was right. Rates of full-time, in-office work plummeted during the pandemic, and while many employers have since shifted from fully remote work to hybrid work schedules, hybrid work levels have remained stable since 2022. In 2025, 78% of full-time remote-capable U.S. employees are either hybrid or fully remote, with hybrid as the most common arrangement.

What is sustaining this transformation of the workplace? The inconvenient answer for those who see government action as the source of progress is that this transformation is thanks to freely chosen, mutually beneficial decisions by employers and employees rather than any top-down mandate. 

The proof is in how durable these patterns have been. Millions of workers and employers continue to choose them because they work. Roughly half of the U.S. workforce consists of employees who are capable of working remote or hybrid jobs. The most common arrangement sees workers commute two days a week and work remotely the other three. Most workers appreciate the flexibility, with a mere 6% of remote-capable workers saying they prefer to work on-site full-time.

An analysis from the Bureau of Labor Statistics found that across industries, the rise in remote work and total factor productivity growth may be positively correlated, and many employees report higher productivity at home. Some research also suggests that hybrid and fully remote work may have positive effects on individual employee productivity, satisfaction, and physical health, as well as employee retention.

Historically, most improvements for workers have followed a similar pattern—wherein the market dictates employer-employee relationships, not government—despite a popular narrative to the contrary. The economist Benjamin Powell observed that legal labor standards, working-hour limits, and the introduction of a minimum wage in the United States and other wealthy countries after industrialization largely mirrored policies that employers had already implemented of their own accord. Legislation merely codified preexisting norms instead of prompting a change in industry practices. 

Economist Price Fishback similarly noted, for example, “State laws limiting the number of working hours for women … passed after many employers had substantially reduced hours for women. Recent studies have found that the laws had relatively little effect.” 

A century ago, the Ford Motor Company pioneered limiting the workweek to five days. Ford’s example soon inspired manufacturers across the country and around the world to adopt the Monday-to-Friday workweek. That occurred because employers discovered that productivity increased, while employees valued the extra leisure time. As the economist Ludwig von Mises put it, “The nineteenth century’s labor legislation by and large achieved nothing more than to provide a legal ratification for changes which the interplay of the market factors had brought about previously.”

Today, remote and hybrid work often benefits both employers and employees, and this work flexibility rise occurred in spite of many outdated government rules that hinder such arrangements. These rules are in desperate need of reform. For example, various federal tax rules discriminate against remote work arrangements, while differing state rules can subject remote workers to double taxation, and occupational licensing rules limit workers’ options to move between states. If anything, the government has stood in the way of the great workplace transformation toward remote and hybrid work.

Calls to legally mandate remote or hybrid work are a misguided attempt to give the government credit for a shift that has already happened independently of government action. Premier Jacinta Allan of Victoria, Australia, announced that her state government will enshrine a legal right for employees, in both the private and public sectors, who can perform their job from home to do so at least two days a week. The law is expected to come into effect in September.

The legal change will benefit few employees, as 65% of Victorians are already hybrid or remote, but it will create more bureaucratic headaches by adding unnecessary red tape for employers and employees alike. “WFH [work from home] is already happening, and there is no reason to legislate a one-size-fits-all approach,” cautioned Andrew McKellar, the chief executive of the Australian Chamber of Commerce and Industry.

This represents perhaps the first example of a legal entitlement to work from home, coming long after the market has already made such arrangements widespread. “If you can do your job from home, we’ll make it your right—because we’re on your side,” said Allan. In reality, it is employers and employees exercising their freedom in the market, not political mandates, that have made the flexibility of remote and hybrid work widely available today.

This article was originally published at RealClearMarkets on 6/23/2026.

Center for AI Safety | Science & Technology

A Significant Increase in Digital Labor Automation

“The Remote Labor Index (RLI) measures how often AI agents can complete real, economically valuable freelance projects (3D & CAD, architecture, graphic design, video and animation, audio, data analysis, web apps, and more) at a quality a paying client would actually accept. Every deliverable is judged by human evaluators against a gold-standard deliverable produced by a paid professional. The headline metric, the automation rate, is the share of projects where the AI’s work is judged as good as (or better than) the human’s.

At the benchmark’s release, the best AI agent automated just 2.5% of projects. Today we’re publishing results for three newer models, paired with stronger agent scaffolding. Automation rate is rapidly increasing…

Fable 5 reaches the highest automation rate measured so far, 16.1%, roughly double Opus 4.8 at 8.3%. GPT‑5.5 reaches 6.3%. All three score above every previously evaluated model.

For context, the previous published leader sat at 4.17% (Opus 4.6 with the Claude Cowork scaffold), and the field topped out at 2.5% when RLI was released. The frontier has more than quadrupled in under eight months, a concrete signal of how quickly economically capable AI agents are advancing.”

From Center for AI Safety.