When businesses talk about AI, productivity is usually one of the first things they mention. Work that took hours can be done in minutes. Smaller teams can produce what once required larger ones. People can research, write, code, analyze and create faster than before. From a business point of view, this makes perfect sense. If ten people can now produce what once required fifteen, the company has become more productive. It can produce more with the same team, or produce the same amount with fewer people.
It is the second possibility that makes the economics of AI more complicated. Technology has been doing this for a very long time. During the Industrial Revolution, machinery allowed factories to produce far more with less human labor. It also raised a question economists were already arguing about: does greater productivity necessarily make the people doing the work better off?
David Ricardo initially believed machinery would generally benefit workers along with everyone else. He later changed his mind. In 1821, he added a chapter called “On Machinery” to the third edition of Principles of Political Economy and Taxation, acknowledging that machinery could increase production while hurting workers whose labor was no longer required.
Karl Marx took the argument much further. For Marx, the machine itself wasn’t really the problem. What mattered was who owned it and what happened to the value it created. If technology allowed a business to produce more with less labor, the benefits of that productivity did not automatically go to the workers in the form of shorter hours, higher wages or greater security.
A software company discovers that AI allows 70 people to do work that previously required 100. It could keep all 100 and produce considerably more. It could reduce working hours. It could use some of the gains to pay people more. Or it could employ 70 people and reduce its costs.
Which option a company chooses depends on how it wants, or needs, to use that productivity. And it isn’t simply a question of whether the people running that business are good or bad. Businesses operate within incentives of their own. Lower costs can mean better margins. Investors expect returns. Competitors become more efficient. A company that continues employing 100 people to produce what its competitor can produce with 70 eventually has to explain why.
AI is changing the economics of how many people a business needs. This isn’t entirely new either. In 1930, John Maynard Keynes imagined that technological progress would eventually make society so productive that people might work only around 15 hours a week. He wrote about what he called “technological unemployment,” but believed it could be a temporary stage on the way to greater abundance and leisure.
Almost a century later, we are becoming extraordinarily good at producing more with less time, yet the prospect of needing less human labor is creating anxiety rather than relief. If technology can do more of our work, why isn’t the obvious outcome that humans get to work less?
This takes me back to Marx’s question about ownership. Who owns the productive technology, who controls it and how are its gains distributed? AI complicates this because access to productive technology is no longer limited to large businesses or people with substantial capital. The technology that can make an employee less necessary to a large company can also make the large company less necessary to the individual.
A developer can build something that once required a larger engineering team. A consultant can research and analyze without employing several people. A designer or marketer can access capabilities that would once have required an agency. A small company can automate functions it could never have afforded to staff. You don’t need to own a factory to access a remarkable amount of productive capacity anymore. Sometimes you need a laptop and a collection of subscriptions.
So AI potentially does two contradictory things. It can concentrate power and decentralize power at the same time. Power can concentrate among the companies that own the models, computing infrastructure, proprietary data and platforms on which everyone else increasingly depends. Established businesses can also capture productivity gains by producing more with fewer employees.
At the same time, productive capability is spreading to people and small businesses that could never previously have afforded it. A large company can look at AI and say, “We need fewer people.” An individual can look at the same technology and say, “I need less of the company.”
That second possibility is easy to overlook when we talk about AI primarily through layoffs and job losses. The employee displaced by technology doesn’t automatically become an entrepreneur, of course. Access to AI doesn’t give everyone capital, customers, expertise or the ability to build a business. But it does lower some of the barriers that once separated an individual from an organization with considerably greater resources.
We still don’t know where this takes us.
Marx gives us one possibility: technology strengthens those who own productive capital and allows them to capture more of the gains from productivity. Keynes gives us another: extraordinary productivity could eventually reduce how much humans need to work. And the entrepreneur with a laptop gives us a possibility neither picture captures particularly well: productive power that once belonged mainly to organizations becoming available to individuals themselves. All three may be happening at once.
We keep asking how many jobs AI will replace. But if AI changes how much labor, capital and organizational infrastructure are required to create economic value, it could change the company itself, not just the jobs inside it.
As we learn to do more with less, the old questions haven’t gone away: who owns it, who controls it, and how are the gains distributed?