This simple question separates technical efficiency from free time. The machine can reduce the effort needed. Whether that becomes a shorter working day, more output, a lower price, a higher profit or a lost job is not something the machine decides.
Why did Keynes's promise of leisure not fulfil itself?
In 1930 John Maynard Keynes imagined that technical progress might solve the economic problem, or at least bring it within sight of solution, within a hundred years. For the work that would still remain, he thought three-hour shifts or a fifteen-hour week conceivable. His horizon ends in 2030. It was less a timetable than a thought experiment: what happens when productivity is no longer our scarcest good?
On one decisive point Keynes was right. Technical systems can dramatically reduce the human effort per item produced. A modern loom, a machine tool or a spreadsheet accomplishes in a short time work that once took many hands and hours. The history of industrialisation is also a history of minutes gained. But those minutes do not afterwards lie beside the machine as a free reserve.
A business can produce more with the same staff. It can cut jobs, lower prices, raise wages or create new products and tasks. Employees can earn more without working less. Owners can keep a larger share of the proceeds. Customers may get a cheaper product, but not a free hour. Each of these possibilities uses the productivity gain differently.
Historically, working hours did fall in many countries while wages and productivity rose. The International Labour Organization points out, however, that in recent decades this long-term decline has in some cases stalled or even reversed. Productivity can make shorter work possible. It does not enforce it.
When efficiency creates new work
As early as 1865, William Stanley Jevons noticed a striking pattern in coal consumption. More efficient steam engines needed less coal for the same output. But because they became cheaper to run, they were used for more and more purposes. Using a resource more economically did not have to reduce total consumption and could even increase it. This observation later became known as the Jevons paradox.
Human labour is not coal, and an observation about energy economics yields no natural law of the labour market. The analogy simply exposes a good question: what happens when a task becomes cheaper and faster? Often it does not simply become rarer. More variants appear, shorter deadlines and expectations that would previously have been too expensive. A faster design tool can free up an afternoon. It can just as easily create the expectation of presenting ten variants by evening.
The economist David Autor explains why automation has historically not simply eliminated all work. Machines replace certain tasks, complement others and create new demand through higher output. Daron Acemoglu and Pascual Restrepo distinguish between a displacement effect, in which capital takes over tasks previously done by people, and a countervailing effect through new tasks in which human labour is needed. Automation can raise productivity and at the same time reduce labour's share of the value created. No single figure on a balance sheet says who has gained time.
What the new AI studies actually measure
With generative AI, the difference can already be seen on a small scale. Shakked Noy and Whitney Zhang had 453 college-educated professionals complete typical writing tasks. Access to ChatGPT cut the average time taken in this limited experiment by 40 percent and raised the rated quality by 18 percent. That is a clear productivity gain. It says nothing yet about whole professions, lasting jobs or free Friday afternoons.
A field study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond examined 5,172 customer-support agents. An AI assistant raised the number of issues resolved per hour by 15 percent on average. Less experienced agents benefited most; the most experienced barely became faster, and their quality even declined slightly. The very measure used shows where the minutes gained went first: into more cases handled per hour, not into a shorter shift.
This is not a hidden refutation of the technology. The same study found evidence of learning and better customer conversations. A tool can make work faster and more bearable at the same time. But the benefit has several dimensions, and they do not automatically go to the same person. The company gains capacity, customers get faster help, employees perhaps less stress or a higher target. Only the organisation of work decides the mix.
A 2026 research review by the International Labour Organization paints a cautious picture. The productivity gains from generative AI are real, but often not yet independently verified and unevenly distributed. Time savings reported by workers, amounting to a few percent of working hours, have not yet translated into higher measured output, earnings or employment. Between technical possibility and social outcome stands an institution.
Who gets the 59 minutes?
A machine does not store human time. The hour saved does not exist as an object. A particular task simply requires less human working time. Economically, that opens up room for manoeuvre, and that room is divided up through competition, ownership, contracts, taxes, collective bargaining and expectations. The real time machine, then, is not the robot. It is the rule that decides what happens after it is put to work.
That can mean shorter working weeks, reliable schedules, a share in the productivity gain or social security when a job disappears. Equally possible are work intensification, constant availability or a split between overworked employees and people whose work is no longer paid for. Both can arise from the same technical efficiency.
With this, the three-part series comes full circle. Free time does not guarantee a good life: the first essay showed that even an unplanned hour can be empty and tormenting. Nor is work merely stolen time: the second essay described how it can carry meaning, skill and community. Yet without control over an hour, there is not even the possibility of filling it oneself.
Robots do not own our time. They merely reveal that productivity and freedom are different quantities. The political question of automation is therefore not only: what can the machine do? It is: who gets the 59 minutes once it has finished?
Sources
John Maynard Keynes, Economic Possibilities for Our Grandchildren (1930)
William Stanley Jevons, The Coal Question (1865)
David H. Autor, Why Are There Still So Many Jobs? The History and Future of Workplace Automation, Journal of Economic Perspectives 29 (2015)
Daron Acemoglu and Pascual Restrepo, Automation and New Tasks: How Technology Displaces and Reinstates Labor, Journal of Economic Perspectives 33 (2019)
Shakked Noy and Whitney Zhang, Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence, Science 381 (2023)
Erik Brynjolfsson, Danielle Li and Lindsey Raymond, Generative AI at Work, The Quarterly Journal of Economics 140 (2025)
International Labour Organization, Working Time and the Future of Work, ILO Research Paper Series (2018)
International Labour Organization, The Impact of GenAI on Jobs, Productivity and Work Organization: A Review of the Empirical Evidence, ILO Research Brief (2026)