In 1930 John Maynard Keynes published a short essay destined to become one of the most famous economic prophecies of the twentieth century: Economic Possibilities for our Grandchildren. In that text, he imagined that, about a century later, technical progress, the increase in productivity, and the accumulation of capital could drastically reduce the working time necessary to satisfy the material needs of society. He went so far as to hypothesize three-hour shifts per day, or a working week of about fifteen hours.
It was not a naive prediction. Keynes did not think that human beings would stop desiring, competing, producing, and accumulating. However, he sensed a historical possibility: if technology continued to increase productivity, the main problem of modern man would no longer be scarcity, but the use of liberated time.
A century later, that prediction appears at least partially betrayed. Not because the technology did not work. On the contrary: it has worked extremely well. Productivity has grown, communications have become instantaneous, access to information has enormously expanded, and many manual, administrative, and cognitive activities have been simplified. Yet we do not work fifteen hours a week. In many cases we work more, are more reachable, more exposed, more solicited, immersed in an operational continuity that invades evenings, weekends, and private spaces.
The question, then, is inevitable: why hasn't technology given us back time?
The digital revolution has given us speed, but has taken away boundaries
Over the last thirty years, the digital revolution has transformed almost every dimension of work. The internet, email, the cloud, video conferencing, instant messaging, and collaborative platforms have reduced distances, transmission times, bureaucratic steps, and many material waiting times. Where before it required travel, physical archives, paper copies, repeated phone calls, and mandatory meetings, today we can send a contract in seconds, attend a meeting from anywhere, share documents, coordinate workgroups, and distribute content in real time.
All of this was supposed to make our lives simpler, and in part it did. But the time saved did not automatically translate into free time. Very often it was reabsorbed into new tasks, new communications, new urgencies, and new expectations of availability. The digital revolution eliminated some wait times but produced permanent reachability. It made it simpler to communicate, but also harder to escape communication. It reduced many idle times but multiplied work opportunities.
The result is paradoxical: tools born to lighten work have often ended up expanding it. We do not necessarily work better; we are only more connected, more reachable, and more solicited. One statistic helps understand the nature of the problem: according to Microsoft's Work Trend Index, in Microsoft 365 applications, the average worker spends 57% of their time on communication activities — meetings, email, chat — and only 43% on the actual creation of documents, spreadsheets, and presentations. The most intensive users reach nearly nine hours a week dedicated to emails and over seven hours to meetings.
This figure is striking because it shows that an enormous part of contemporary work does not consist of the direct production of value, but of the management of work itself: coordinating, responding, searching, realigning, convening, summarizing, updating, chasing. Modern work has become, to a large extent, work on work.
Generative AI introduces a different discontinuity
Generative artificial intelligence changes the question. Not because it is a magic wand or because it can automatically replace skills, judgment, responsibility, and professional sensitivity. But because it intervenes in a part of work that the digital revolution had only accelerated: intermediate cognitive work.
Much of contemporary profession does not consist only of making final decisions, but of preparing the ground for those decisions to be possible: reading documents, ordering information, summarizing materials, comparing alternatives, transforming notes into texts, preparing drafts, building presentations, reconstructing regulatory frameworks, making an analysis communicable. It is this intermediate zone, often tedious and dispersive, that generative AI can significantly compress.
In many high-intensity cognitive professions — lawyers, accountants, engineers, architects, consultants, managers, researchers, teachers, communicators — the leap can be of a different order of magnitude compared to simple digital automation. A report that previously took days can be drafted in a few hours; a presentation that once took a day can be structured in much less time; a preliminary research that previously took weeks can be oriented and organized in enormously shorter times. A complex analysis can be accompanied, verified, reformulated, and made communicable with a speed previously unthinkable.
The general metrics already available are significant, but often conservative. A study by the Federal Reserve Bank of St. Louis estimated that users of generative AI save an average of 5.4% of hours worked, which is about 2.2 hours a week on a 40-hour week. The Harvard Business School/BCG study on consultants found that professionals with access to GPT-4 completed 12.2% more tasks, worked 25.1% faster, and produced qualitatively better results in a significant portion of activities. McKinsey estimated that generative AI and related technologies could automate activities that today absorb between 60% and 70% of employees' working time.
These are important numbers. But, in my view, they risk underestimating the impact on professional jobs most exposed to the production of texts, analyses, reports, opinions, documents, presentations, research, and summaries. Averages measure the average use of AI, not deep use. It is one thing to use it to correct an email; it is another to use it to draft a technical report, compare complex documents, transform scattered materials into a coherent structure, analyze data, or prepare a draft to be submitted for professional verification. In the first case, AI saves minutes. In the second, it can save days.
From executive work to managerial work
This transformation does not mean that human competence becomes useless. The opposite happens. The more powerful the tool becomes, the more decisive the judgment of the user becomes. AI can produce a draft, propose a structure, summarize a document, build an analysis, or generate alternatives; but it remains human to understand if that draft is correct, if that structure makes sense, if that summary omits something, if that analysis holds up, and which alternative deserves to be chosen.
Work does not disappear: it shifts. It shifts from material writing to directing the process, from compilation to validation, from repetitive execution to critical verification, from producing the first version to responsibility for the final version. In this sense, AI does not eliminate the professional; it eliminates, or drastically reduces, a portion of the effort that often prevented them from dedicating themselves to the highest part of their work.
For an engineer, the point is not just to write a report faster, but to have more time to think about the setup, technical coherence, the quality of assumptions, the clarity of presentation, and the responsibility of the conclusions. For a lawyer, it means being able to dedicate more attention to strategy, argument, risk, precedents, and interpretation. For an accountant, it means shifting from compilation to advice, planning, and control. For an architect, it means gaining time for concept, proportions, context, function, and beauty. For a researcher, it means reading and organizing literature faster to formulate better questions.
True competence will be less and less about doing everything manually and more and more about directing, questioning, verifying, integrating, interpreting, and deciding.
The next leap: AI agents
The most profound transformation, however, is yet to arrive fully: that of AI agents. So far, we have used artificial intelligence primarily as an assistant: we ask it for a draft, a summary, a translation, a revision, a comparison, an outline, a preliminary analysis. Agents represent a further step, because they do not just answer a question. They can receive a goal, break it down into sub-activities, use tools, read documents, query databases, produce intermediate outputs, update files, prepare reports, interact with different software, and construct an operational sequence.
It is no longer just the AI that writes. It is the AI that executes parts of a process. This distinction is fundamental. The digital revolution accelerated single actions like sending, searching, archiving, and communicating. Generative AI accelerates single cognitive activities like writing, summarizing, translating, comparing, and reformulating. AI agents can accelerate entire workflows.
This is where the impact on professions, businesses, and public administrations could become enormous. A lawyer will be able to use agents to analyze files, build timelines, compare contracts, search case law, and prepare drafts. An accountant will be able to automate document checks, schedules, reconciliations, and preliminary tax reports. An engineer will be able to have agents read specifications, regulations, computations, reports, monitoring data, and design drawings, generating preliminary checks, diagrams, and comparative analyses. A manager will be able to entrust agents with the preparation of dashboards, meeting summaries, follow-ups, market analyses, and budget scenarios. A researcher will be able to use them to monitor literature, compare papers, build bibliographical maps, identify scientific gaps, and organize data and materials.
According to Gartner, by 2026, up to 40% of enterprise applications could integrate task-specific AI agents, compared to less than 5% in 2025. Deloitte predicted that 25% of enterprises using generative AI would adopt agents in 2025, with growth up to 50% by 2027. We are no longer in the realm of science fiction: we are facing an industrial transformation already underway.
Of course, AI agents are not yet mature for everything. They can make mistakes, fabricate information, misunderstand goals, access sensitive data, and generate organizational and legal risks. Gartner itself reported the risk that over 40% of agentic AI projects will be cancelled by 2027 due to costs, low value, or insufficient controls. But this caution does not reduce the scope of the transformation: it clarifies it. Agents must not be imagined as uncontrolled replacements for humans, but as systems to be governed with supervision, authorizations, traceability, verification, responsibility, and differentiated levels of autonomy. Autonomy without control would be dangerous; control without autonomy would make the revolution useless. The real challenge will be to design a new balance.
Who will take the liberated time?
We thus arrive at the central point. If generative artificial intelligence can save us hours, and AI agents can compress entire processes, what will we do with the freed-up time? Will we use it to work less or to produce even more? Will we return it to the person or hand it over to the market? Will it become free time or a new mandatory availability?
This is the question that Keynes had glimpsed nearly a century ago. Technology can enormously increase productivity, but it does not decide on its own how to distribute its benefits. It can free up time, but it does not guarantee that that time remains free. It can reduce fatigue, but it can also raise performance standards. It can make work faster, but it can also multiply expectations.
It has already happened with digital. Email did not reduce the number of communications: it multiplied them. The online meeting did not necessarily reduce meetings: it often increased them. The ability to work anywhere did not always improve individual freedom: sometimes it transformed every place into a potential office. The risk is that the same will happen with AI: if a report can be produced in a tenth of the time, the professional might be asked to produce ten reports; if a case can be managed faster, the number of cases might be increased; if a manager can receive continuous reports, the demand for continuous updates might grow.
But with agentic AI, a different phenomenon compared to the digital revolution could also emerge. Digital accelerated communication, archiving, and coordination, but often left the overall structure of processes intact. AI agents, on the other hand, can impact the entire sequence of work: receiving an objective, breaking it down into tasks, consulting documents, using tools, producing intermediate outputs, updating files, preparing reports, and providing verifiable versions. If this capability truly matures, in many sectors it will not just be about doing the same thing faster, but about drastically compressing entire operational cycles.
This means that, at least for some professions and some organizations, there might not be enough new demand to fill all the freed-up time. There will not always be enough new clients, new cases, new projects, or new activities to absorb the increase in productivity. A professional who takes a tenth of the time to prepare a first draft will not necessarily find ten times more clients. A technical studio that automates significant parts of analyses, reports, and presentations will not automatically get ten times more assignments. An office that drastically reduces processing times will not always have a proportionally larger volume of proceedings.
This possibility changes the nature of the problem. In the digital revolution, the time saved was often reabsorbed by new communications, new meetings, new urgencies, and new reachability. In agentic AI, however, a portion of the freed-up time might not be reabsorbable in the same way, because the compression of processes could exceed the growth of demand. At that point, the question will no longer be "how to work more thanks to AI?", but "how to redistribute the value and time produced by AI?".
The time freed by technology does not automatically become human time. It can be captured by the organization, absorbed by the market, transformed into new expectations of clients, converted into new bureaucracy, used to raise productivity standards, or to generate new forms of control. But it can also open up a real space to reduce workload, rethink schedules, improve the quality of services, dedicate more time to verification, relationship, training, creativity, and life. The real issue, therefore, is not just technological. It is cultural, social, economic, organizational, and political.
Who should start thinking about freed-up time?
A transformation of this magnitude cannot be left solely to the decisions of businesses, market dynamics, or the individual choices of workers. If artificial intelligence can profoundly alter the quantity and quality of work required, then the distribution of freed-up time becomes a public issue. It concerns the model of society we want to build, the relationship between productivity and well-being, the distribution of wealth, the organization of cities, the education of people, and the very meaning attributed to work.
Politics should start asking questions today, before the transformations become difficult to govern. Not only to regulate AI risks, protect data, or defend the most exposed professions, but to understand how to distribute the benefits of increased productivity. The issue is not just about how many jobs will be replaced or transformed. It also concerns the length of the working week, the right to disconnect, lifelong learning, the protection of professional transitions, workers' participation in the benefits of automation, and the possibility that a portion of the wealth produced by technology is converted into greater personal and collective freedom.
Politics, in other words, should not limit itself to governing the potential loss of work. It must also learn to govern the potential gain in time. This is a more difficult task, as it requires a vision that goes beyond the emergency, beyond the life of a single legislature, and beyond the simple measurement of economic growth. It means asking whether the increase in productivity should translate exclusively into higher profits and greater production volumes, or also into shorter working days, better services, greater access to culture, more time for care, and a different quality of life.
In this reflection, the university should also assume a central role, alongside politics, businesses, social organizations, and cultural institutions. Its contribution cannot be limited to the study of algorithms, the training of specialists, or the introduction of AI in degree courses. The university should also help society understand how the increase in productivity can transform work, the distribution of value, and the use of freed-up time.
It can do this through interdisciplinary research, fostering dialogue among economists, engineers, jurists, sociologists, philosophers, psychologists, doctors, planners, and organizational scholars; but also through the training of new generations, the experimentation of new organizational models, the evaluation of the social effects of automation, and the construction of public spaces for debate.
Freed-up time, in fact, is not an exclusively economic or technological issue. It involves identity, health, relationships, culture, education, democracy, and the way people attribute meaning to their lives. For this very reason, the university can contribute not only to describing the change, but also to developing criteria, scenarios, and proposals capable of guiding it.
An equally important task belongs to foundations, independent research centers, cultural associations, trade unions, third-sector organizations, and all those realities that provano to think about the future before it becomes present. These organizations can create spaces for debate that are less conditioned by productive urgency and the electoral cycle, bring together different skills, promote experimentation, and keep open questions that the market naturally tends to reduce to issues of efficiency and profit.
We need permanent observatories on freed-up time, not just on lost employment. We need local experimentation, agreements between companies and workers, university laboratories, urban policies, and cultural initiatives capable of verifying in practice what happens when a portion of the saved time is returned to people. We could measure not only how many working hours are eliminated by AI, but where those hours end up: whether they become new activities, greater pressure, unemployment, training, care, civic participation, or genuine free time.
Businesses must also assume responsibility. It would be short-sighted to consider artificial intelligence solely as a tool to reduce costs or increase the number of services. A portion of productivity gains could be invested in reducing working hours, in the quality of work, in training, in safety, in creativity, and in the opportunity to dedicate more attention to truly important decisions. An intelligent organization should not limit itself to asking every worker to produce more thanks to AI. It should ask how to use AI to build a more sustainable, competent, and human way of working.
The future of freed-up time will therefore depend on a collective choice. If politics, universities, businesses, and social organizations do not start discussing it, a portion of the saved time could be absorbed by existing structures: higher production, new services, additional consumption, higher expectations, and more pervasive forms of control.
However, it is not certain that this reabsorption can occur everywhere and to an unlimited extent. If, in some sectors, artificial intelligence and autonomous agents produce productivity increases of an order of magnitude, the growth in demand may not be sufficient to occupy all the time made available. It will not always be possible to proportionally multiply clients, practices, projects, relationships, or decisions.
Very different situations could therefore emerge. In some sectors, freed-up time will likely be filled by new tasks and new performance standards; in others, it could translate into staff reductions, concentration of economic value, greater precariousness, or an actual reduction in working hours. It is precisely this plurality of outcomes that makes preventive reflection necessary.
If the issue is not addressed consciously, the benefits of increased productivity risk being distributed unequally: some may gain time and autonomy, others may face greater pressure or lose part of their professional role. If, on the other hand, the change is governed, artificial intelligence could become the opportunity to rethink not only work, but the social contract that binds productivity, wealth, rights, and life.
From productive time to human time
For this reason, I believe that artificial intelligence should be discussed not only in terms of efficiency, competitiveness, and automation, but also in terms of personal time, sociality, culture, beauty, education, care, and relationships. This discussion cannot remain confined to technology companies or individual professionals. It must enter universities, institutions, parliaments, trade unions, cultural organizations, and the places where a society tries to imagine its future.
If AI truly allows us to compress activities from days to hours, from hours to minutes, we must have the courage to ask ourselves whether we simply want to fill that space with more work or return it to life. And we must build the political, economic, and organizational conditions so that this return does not remain only a theoretical possibility reserved for a few.
Freed-up time could become time for reading, music, sports, hobbies, nurturing relationships, authentic learning, slow thinking, civic participation, and even the possibility of not always being productive. The highest promise of artificial intelligence should not be to transform us into more efficient workers twenty-four hours a day, but to allow us to be more human: not less competent, but less crushed by repetition; not less responsible, but less consumed by execution; not less productive, but freer to choose where to place our productivity.
Reopening Keynes's prophecy
Perhaps Keynes got the timing wrong, not the question. We do not yet live in a society of fifteen-hour weeks. On the contrary, many live in a society of continuous work, permanent availability, measurable performance, and relentless connection. But the arrival of artificial intelligence, and especially AI agents, reopens that question in a new way.
For the first time, a technology does not limit itself to accelerating communication or access to information. It intervenes directly in the cognitive, organizational, and professional activities that absorb an enormous part of our days. It can compress the time of writing, research, synthesis, preparation, revision, coordination, and intermediate production. But it cannot decide for us what to do with the freed-up time.
This decision remains human. But it cannot be solely individual. The individual worker can hardly defend the saved time alone if the organization, the market, and the dominant culture immediately transform every increase in efficiency into a new demand for productivity. For this reason, the management of freed-up time must become a political, academic, and social issue.
We must build institutions capable not only of accompanying innovation, but also of guiding it; universities capable not only of teaching the use of AI, but of questioning its purposes; organizations capable of imagining the future before it is determined exclusively by economic and technological forces.
And perhaps this will be the great challenge of the coming years: not only learning to use artificial intelligence, but learning not to allow the time it could return to us to be taken away again. The digital revolution taught us to be always connected. Artificial intelligence should help us to be freer.
The real challenge will not be using AI to work more and more. It will be preventing the freed-up time from being colonized once again by work. And it will be necessary to start discussing it now, before the new rules of work are defined without collective reflection. Because technological progress becomes truly human progress only when it returns time to the person, the community, beauty, knowledge, and life.
Essential sources
Keynes, J. M. (1930), Economic Possibilities for our Grandchildren.
Microsoft (2023), Work Trend Index: Will AI Fix Work?.
Federal Reserve Bank of St. Louis, study on average time savings associated with the use of generative AI.
Dell’Acqua, F. et al. (2023), “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality”, Harvard Business School / Boston Consulting Group.
McKinsey Global Institute (2023), The Economic Potential of Generative AI: The Next Productivity Frontier.
Gartner (2025), predictions on the integration of AI agents in enterprise applications and warnings on the risk of cancellation of agentic AI projects by 2027.
Deloitte (2025), Technology, Media & Telecommunications Predictions, predictions on the adoption of AI agents in enterprises using generative AI.
Note on the use of artificial intelligence
This article was prepared with the support of artificial intelligence tools for research, analysis, organization of sources, and writing. The selection of content, critical interpretation, and final responsibility for the text remain with the author.