AI Doubles Your Productivity—Will It Double Your Salary?

When a tool doubles output, where the extra gains flow is never decided by the tool itself.

"AI can help you do twice the work." This line has become almost the most popular narrative in today's workplace. But the second half of the sentence is often conveniently omitted: who ever said you'd get paid twice as much for it?

A short commentary from YouTube puts this question front and center. It doesn't discuss model parameters, nor does it predict job gains or losses. Instead, it uses historical experience from two hundred years ago to force us to answer a more basic question: after productivity rises, do ordinary people's lives actually get better?

The Historical Reference Answer: Output Rose, Wages Lagged

The video cites a commonly used set of estimates by economist Robert Allen: between 1780 and 1840, during the British Industrial Revolution, per capita output rose by about 46%, while real wages—adjusted for inflation—grew only 12%. Machines did make workers produce more, but it took far longer for pay to catch up.

It should be noted that the precise magnitude of this gap remains disputed among historians. The video also acknowledges that "Historians debate the size of that gap"—different studies give inconsistent answers about exactly how wide the gap between output and wages was.

More importantly, the industrial environment of two hundred years ago cannot be directly compared to today's knowledge work. Treating the Industrial Revolution as a prophecy for the AI era is itself a simplification. But the value of this history lies not in precise replication, but in revealing a structural fact: technology enhances production capacity, while distributing the fruits of production is an entirely separate mechanism.

Robert Allen is an economic historian at Oxford University who systematically estimated wage and output data from the Industrial Revolution in works such as The British Industrial Revolution in Global Perspective. His core methodology is the "subsistence wage basket": using the bundle of goods needed to maintain basic survival as a benchmark, and converting nominal wages into real purchasing power based on its price. Allen's estimates show that from the late 18th century to the mid-19th century, British workers' real wage growth lagged far behind per capita output growth—a finding known as "Engels' Pause." It's worth noting that Allen's data are mainly based on wage records of urban manual laborers such as construction workers, with limited coverage of rural and textile industries. Therefore, different scholars using alternative data have reached divergent conclusions, which is the source of the video's statement that "historians still debate this."

After Productivity Doubles, the Money Could Flow Three Ways

The video uses a very concrete thought experiment: suppose AI helps you compress a week's work into three days. What happens next?

First possibility: the company gives you a raise. Second possibility: salary stays the same, but you're loaded with more tasks. Third possibility: the company chooses to cut prices to grab market share, and the extra efficiency becomes a price advantage rather than your income.

None of these three paths is "natural"—they depend on bargaining power, industry competition dynamics, labor market supply and demand, and the company's orientation toward profit distribution. The tool itself is neutral; it only makes the pie bigger. How to slice it is a human decision.

This is also the weakest point of claims like "AI makes everyone more efficient." It describes a fact on the production side while sidestepping the question on the distribution side. And for most people, the latter is the variable that truly affects their lives.

"Bargaining power" is a core concept in labor economics, referring to the influence employees can exert when negotiating pay and working conditions with employers. It depends on multiple factors: the tightness of the labor market (employees have weak bargaining power when unemployment is high), union coverage and collective bargaining systems, the irreplaceability of skills, and employees' exit options (i.e., the ease and cost of changing jobs). In the AI context, bargaining power is also affected by a new variable: the extent to which tools can replace human labor in a given role. If AI makes a skill easily replicable, workers possessing that skill lose bargaining power; conversely, if AI amplifies the leverage of scarce skills, workers may gain a stronger negotiating position. The divergence in distribution outcomes is largely the result of divergent bargaining power.

"More Efficient" Doesn't Mean "Easier"

Office automation over the past few decades has already provided some reference points. Computers, email, and collaboration software have dramatically increased information processing speed, but many white-collar workers' hours haven't shortened as a result. Instead, because response speeds are faster and more tasks can be handled, boundaries keep getting pushed back. Efficiency tools are often first used to take on more work rather than being converted into leisure.

AI's situation is similar. When one person with tools can cover output that previously required multiple people collaborating, companies' more likely short-term response is to adjust staffing and task loads rather than simultaneously raising everyone's pay. Real change often only becomes apparent after the labor market, organizational forms, and bargaining mechanisms all adjust together.

This isn't to say AI necessarily makes things worse. Historical experience also shows that output growth ultimately transmits to living standards in some form—it's just that transmission takes time, and the path isn't automatic.

"Office automation" is not a new phenomenon. Since the 1980s, personal computers, email, and ERP systems have gradually replaced processes like typing, filing, and paper-based approvals, and per capita information processing volume among white-collar workers has risen substantially. Economics has an observation about this called the "Solow Paradox"—in 1987, Robert Solow said, "You can see the computer age everywhere but in the productivity statistics." This paradox was later partly explained as: the benefits of new technology require complementary organizational process reengineering to be realized, and companies often first do "take on more tasks with the same organizational structure" rather than reducing work hours. AI's implementation path will likely follow similar logic: before organizational forms and management systems make corresponding adjustments, efficiency gains are more likely to translate into increased task volume rather than increased leisure.

The Question Has Changed

The question posed at the end of the video is worth everyone's consideration: when someone says AI will make everyone more productive, that doesn't answer what people really care about—will your life get better, or are you just expected to complete more?

For individuals, what we can perhaps do is pull the discussion from "efficiency" back to "terms of exchange": in what form does the extra output return to your hands? Is it pay, shorter working hours, stronger workplace voice, or merely a longer to-do list? These questions have no standard answers, but at the very least they should be explicitly asked rather than glossed over by the "efficiency improvement" narrative.

Technology changes possibilities; distribution determines outcomes. Two hundred years later, this principle hasn't changed.