The Four Industrial Revolutions

In February 2026, economists published a survey of nearly 6,000 senior executives in the United States, the United Kingdom, Germany and Australia. Sixty-nine percent of their firms were actively using AI. Yet more than 80 percent of firms reported no impact on either productivity or employment over the previous three years. Looking ahead, executives expected their own firm’s productivity to be about 1.4 percent higher after three more years: a total, not a yearly rate¹.

If that sounds like a disappointment, it should also sound familiar. In 1987 the Nobel economist Robert Solow wrote: “You can see the computer age everywhere but in the productivity statistics”². Companies had been buying computers for two decades, and the economy had little to show for it. A decade later, productivity was surging.

The previous chapter told one version of this story, the electric motors that sat bolted to old line shafts for close to forty years. This chapter steps back to the longer history. The aim is not to make you a historian. It is to give you the pattern, because the pattern is one of the better guides we have to what AI will demand of leaders.

Four revolutions, or five?

The phrase industrial revolution is used loosely, so it helps to be precise. A widely quoted framing comes from Klaus Schwab of the World Economic Forum, who counts four³.

Two details are commonly misquoted. First, Schwab’s Fourth Industrial Revolution is not “the AI revolution”. He describes it as a fusion of technologies that blurs the lines between the physical, digital and biological spheres. AI is one driver among several, alongside robotics, sensors, biotechnology and new materials. Second, the digital revolution did not wait for AI to make computers useful for thinking work. Computers have amplified calculation, record-keeping and analysis since the 1960s. What is new is how far automation now reaches into language- and judgment-heavy work, which is the subject of a later section.

Not everyone counts four. The economist Carlota Perez counts five techno-economic revolutions since 1771, each dated from a visible “big bang”, from Arkwright’s mill in 1771 to the microprocessor in 1971⁴. Whether AI counts as part of the information revolution or the start of a sixth is genuinely debated, and you do not need to settle it. Counting matters less than recognizing what the revolutions have in common.

What makes a technology revolutionary

Most new technologies improve one industry. A few reshape the whole economy. Economists call these general-purpose technologies, and Timothy Bresnahan and Manuel Trajtenberg gave the classic test: steam, the electric motor and the computer all had three properties⁵.

The third property is the one leaders should underline. A general-purpose technology pays off through the complementary inventions that grow around it: the new factory layout around the electric motor, the new supply chains around the railway, the new business processes around the computer. Those complements take years to invent and install. That is why the payoff lags.

For AI, the complements are easy to list and slow to build: redesigned workflows, clean data that authorized systems can reach, new review steps, and people trained to judge what the system produces. None of them comes in the box. This is also why the technology is the smaller part of the bill. A license can be signed in a week. A new way of working takes as long as it always has, because it depends on people changing habits, managers changing measures and controls being designed and tested. An executive who budgets only for the technology has paid for the motor and left the old line shaft in place.

Does AI meet the test? On pervasiveness, the best evidence comes from a study whose title makes the point, GPTs are GPTs. The authors mapped AI capabilities against the tasks in US occupations. They estimated that about 80 percent of US workers have at least 10 percent of their tasks exposed to large language models, and about 19 percent have at least half of their tasks exposed⁶.

Exposed does not mean automated. It means a meaningful part of the task could be done faster or differently with AI. Whether that happens, and who benefits, depends on the complements, which is to say on decisions organizations have not yet made.

The pattern that repeats

Perez’s most useful contribution for executives is not the count of revolutions but the shape of each one. Every revolution she studied went through two long periods with a turbulent break between them⁴.

In the installation period, the new technology attracts a flood of investment and expectation. Infrastructure is built quickly, often more than is needed, and financial speculation runs ahead of real use. The railway manias of the nineteenth century and the internet boom of the late 1990s are the textbook examples. A turning point follows, often a crash, when expectations reset. Then comes deployment: the technology becomes ordinary, organizations redesign around it, and the broad productivity gains finally spread through the economy.

The lesson is not to predict the next crash. It is to keep two things apart that the installation period always blurs: the financial story of a technology and its business value to your organization. Spending more during a frenzy does not buy you the deployment-era gains. Reorganizing does.

The productivity paradox, every time

Each general-purpose technology has produced the same disappointment, for the same reason.

For electricity, as the previous chapter showed, the gains came in the 1920s, when factories were rebuilt around small motors on each machine⁷. For computers, the firm-level evidence is just as clear. When Erik Brynjolfsson and Lorin Hitt studied how companies actually used information technology, the returns depended on what came with it. Firms that paired computers with decentralized decisions, new work practices and investment in their people got far more from the same spending than firms that simply automated what they already did⁸. The national productivity revival of the late 1990s followed years of exactly that kind of reorganization.

For AI, we are early. The 2026 executive survey that opened this chapter is what the early years of a general-purpose technology usually look like: wide adoption, light use and little measured impact. That is not proof AI will disappoint. It is a reminder that the gains will go, as before, to organizations that build the complements.

Story: the line that changed the factory

One of the clearest examples of a general-purpose technology paying off is not a power station or a computer. It is a factory in Highland Park, Michigan.

Ford had been building the Model T since 1908. Electric motors, which let machines be placed where the work needed them rather than along a drive shaft, made a new layout possible. From 1913 Ford’s engineers used it to build the moving assembly line. The car came to the worker, and each worker did one step. The time to assemble a Model T chassis fell from about 12 and a half hours to about 93 minutes⁹.

The second number is the part of the story that is usually left out. The new work was relentless, and people quit. In 1913 Ford hired more than 50,000 workers to keep an average workforce of about 13,600, a turnover rate of 370 percent. In January 1914 the company doubled pay to five dollars a day and shortened the shift. Turnover fell to 54 percent in 1914 and 16 percent the following year [@henryford-five-dollar-day; @raff-summers-1987]. The technology made the new factory possible. The redesign of the work and of the deal with the people doing it made the new factory succeed.

What is genuinely different this time

History is a guide, not a forecast. Three things about AI are genuinely new, and leaders should take them seriously.

The kind of work. Earlier waves automated physical work and then routine information work. Today’s AI can do tasks that were long thought to need a person: drafting, summarizing, reasoning over documents and holding a conversation. That is why exposure is so broad, and why it reaches well-paid professional roles, not only routine ones⁶. It is a statement about capability. As the survey that opened this chapter shows, measured results have not yet followed.

The infrastructure. Electricity needed power stations, wiring and new factories before ordinary firms could use it. AI spreads through computers, networks and cloud services that most organizations already have. The building happens elsewhere, in data centers and chips, which is part of the next chapter’s story.

The interface. Earlier software had to be learned. AI can be instructed in ordinary language, so almost anyone can start using it on the first day. That lowers the cost of trying it, and it also lowers the visibility of how it is being used.

None of these changes the deeper pattern. Value still depends on complements, gains still lag adoption, and the transition still falls on people. They mostly change the speed: the installation can happen faster than before, which leaves less time to work out the redesign.

The transition is never free

Every revolution produced enormous long-run gains, and every one was hard on many of the people who lived through it. In Britain’s early industrialization, output per worker rose for decades while real wages barely moved¹⁰. Ford’s turnover crisis is a small, sharp version of the same lesson. Treating the people side of a technology transition as an afterthought has never worked. It is why the redesign has to cover roles, skills and incentives, not only the process. How AI changes jobs and skills is the subject of AI and the Future of Work, later in this module.

What this means for leaders

The history yields four practical lessons.

1. Plan for the lag. Expect a period in which adoption is wide and measured impact is small. Budget and report for it honestly, rather than declaring early victory or early failure.
2. Invest in the complements. The technology is the smaller part of the cost of a general-purpose technology. The larger part is process redesign, skills, data and new ways of making decisions, and that is where the advantage is.
3. Keep the frenzy and the value apart. Market excitement is not evidence of value in your organization. Judge AI investments by the work they change, not by the attention they get.
4. Redesign the deal, not only the work. People absorb the cost of every transition. The organizations that succeed redesign roles, skills and incentives along with the process.

Check yourself

1. AI is the Fourth Industrial Revolution.MythFact
2. AI is the first technology to amplify thinking work.MythFact
3. In past revolutions, productivity gains followed soon after firms adopted the technology.MythFact
4. Firms that paired computers with new work practices got more out of the same technology.MythFact
5. About four in five US workers have at least a tenth of their tasks exposed to AI.MythFact
6. The gains from past revolutions reached ordinary workers quickly.MythFact

Reflection: find the old shaft

What comes next

Every revolution needed a catalyst, and every one needed a redesign. If AI follows the pattern, an obvious question remains: why now? The ideas behind AI are decades old. The next chapter, Why AI, Why Now?, explains the forces that came together to make it practical in the last few years.