9. Whose Work, Whose Reward?

An hour saved belongs to someone

Imagine a small business that spends every Friday afternoon preparing invoices, checking stock, answering routine inquiries, and assembling a sales report. Its owner introduces an AI assistant that makes the work easier. Errors decline. Staff leave on time. Customers receive quicker answers. The business can serve people it previously had to turn away. This is an invented example, but it captures a benefit worth wanting.

Now ask what happens to the time saved. The owner might shorten the week without reducing pay. The business might sell more, hire someone for a new service, lower prices, increase wages, or retain the money to survive a difficult season. It might also reduce staff. The technology expands the range of possible decisions. It does not make the distributional decision itself.

This is why a discussion of AI and work cannot end with productivity. Producing more value from a given amount of effort can make people better off. It can also produce conflict over who receives the gain, who bears the transition, and who has a say. The productivity question concerns what becomes possible. The distribution question concerns how people share that possibility.

Pope Leo XIV placed this connection near the beginning of his pontificate. Explaining his choice of name to the College of Cardinals on May 10, 2025, he linked Leo XIII’s response to an earlier industrial revolution with contemporary challenges involving AI, human dignity, justice, and labor.¹ The connection invites an inquiry into social arrangements. A transformation in productive power also changes the conditions under which people seek security and recognition.

A useful starting point is to welcome relief from unnecessary drudgery. People do not become more dignified because a bad system forces them to copy information between incompatible forms. A worker should be able to value their contribution while wanting parts of the job to become easier. Defending human worth does not require defending every task that humans currently perform.

We should be equally willing to imagine larger gains. A small enterprise could obtain capabilities previously affordable only to a much larger firm. Someone with limited time could organize a business alongside caring responsibilities. A worker whose strongest ideas are harder to express in formal writing could receive help presenting them. These are possibilities, not guarantees. Their value is sufficient reason to investigate and build useful systems.

The supplied Huang interview transcript presents technology as a source of opportunity, including work involved in constructing infrastructure, and expresses an ambition to reward employees generously.² Those aspirations belong in a fair account of the debate. They do not establish that every worker displaced in one place will obtain a good new job elsewhere. Neither industrial enthusiasm nor industrial fear can substitute for examining the actual arrangements.

The question is therefore more demanding than whether AI creates or destroys work in the abstract. What kinds of work become easier? Which skills gain value? Who must absorb an income interruption? What new services become possible? How does a worker participate in deciding the future of their own workplace? An answer that counts jobs while ignoring these questions may miss much of what makes a working life bearable or worthwhile.

Jobs contain more than tasks

A job is a bundle of activities held together by an institution and a purpose. A receptionist may answer familiar questions, calm an upset visitor, recognize an unusual problem, keep a queue moving, and know when to interrupt someone senior. Some activities may be easy to automate. Others depend on local knowledge, physical presence, discretion, or a relationship built over time. Counting the first set does not settle the fate of the whole job.

The International Labour Organization’s 2025 analysis estimated that one in four workers worldwide was in an occupation with some degree of exposure to generative AI. It explicitly distinguished potential exposure from actual job losses.³ Exposure describes where technology may affect tasks. Employment outcomes also depend on adoption, costs, demand, organization, and choices made by employers and governments.

The distinction prevents two opposite mistakes. We should not turn every exposed task into an announced redundancy. We should not assume that retaining a few human tasks guarantees a stable job for everyone currently doing the work. An organization can combine the remaining activities into fewer positions. Alternatively, lower costs can make it worthwhile to expand a service. Both are possible, and neither follows from a task demonstration alone.

An invented translation business makes the issue concrete. A tool sharply reduces the time needed to produce an initial draft. One owner treats this as a reason to dismiss translators and sell cheaper work with minimal review. Another offers more languages, uses human specialists for difficult material, and serves clients who could not previously afford the service. A third finds that demand hardly grows and cannot sustain its previous staffing. The same broad capability enters three different economic situations.

Measured benefits already provide reasons to take assistance seriously. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond’s study of customer-support work, published in the Quarterly Journal of Economics in 2025, found a 15 percent average increase in issues resolved per hour after access to a generative AI assistant, with substantial differences between workers.⁴ This is evidence about a particular kind of work in a particular setting. It is not a forecast that every occupation will become 15 percent more productive.

In a separate randomized experiment reported in Science in 2023, Shakked Noy and Whitney Zhang assigned professional writing tasks to 453 college-educated participants. Access to ChatGPT reduced average completion time by 40 percent and increased assessed output quality by 18 percent.⁵ Again, the result concerns specified tasks and measures. It shows that useful gains are possible without telling us what happens to an entire labor market after firms reorganize around them.

The scope of a study is not an inconvenience to be removed when communicating it. Scope is what allows evidence to support a decision. A manager choosing a tool needs to know whether the study resembles the work in their organization. A policymaker assessing employment needs evidence beyond short tasks. A worker considering training needs to know which activities the tool handles well and which still require substantial judgment.

There may also be new work in deciding how to use the tool. Someone must select appropriate inputs, inspect outputs, manage exceptions, protect information, and maintain the service. Sometimes these responsibilities create worthwhile roles. Sometimes they are added to existing jobs without adequate time or pay. The appearance of a new task is not automatically the creation of a good job.

For that reason, the unit of concern should remain the working person within an actual institution. Productivity can rise while discretion falls. Pay can rise while hours become less predictable. A role can survive while its path toward advancement disappears. We need enough detail to see these changes separately before deciding whether the transformation deserves to be called progress.

Scroll sideways to see all columns. Keyboard: focus the table and use the left and right arrow keys.

Text equivalent: Gains differ. Sharing them is a decision.

Part of figure	Value or meaning
Title	Gains differ. Sharing them is a decision.
Setting	One customer-support workplace
Lowest fifth	Lowest pre-tool performance fifth Largest gain in issues resolved per hour.
Highest fifth	Highest pre-tool performance fifth No statistically clear gain in issues resolved per hour. That is not proof of zero effect.
Grouping	Groups reflect recent job performance: speed, resolution and customer satisfaction.
Editorial question	A separate decision Who shares the gains through pay, time or better working conditions?
Unmeasured outcomes	The study did not observe wage changes or overall demand for workers.
Diagram type	Conceptual aid • no numerical scale
Source	Source: Brynjolfsson, Li & Raymond (2025), pp. 911–912; 892–893.
How to read the figure	Only the lowest and highest fifths are highlighted. No intermediate groups, coefficients or confidence intervals are estimated here. This study does not establish economy-wide wage or employment effects.

Figure 2. A conceptual reading aid based on the published study, not a coefficient chart. Card dimensions do not show effect sizes. The distribution of productivity gains does not settle pay or hiring.

Productivity does not write a pay agreement

Suppose a tool allows a hypothetical team to produce the same output in four days that previously required five. The arithmetic creates room for several outcomes. The team could keep its hours and expand output. It could work fewer hours. It could serve a more demanding market. Its employer could reduce prices, increase profit, or change staffing. There is no instruction hidden inside the software that selects the morally correct combination.

People sometimes speak as though gains automatically trickle toward every contributor once a business becomes more productive. Others speak as though every gain must inevitably be captured by owners. Both formulations are too complete. Distribution depends on bargaining power, competition, ownership, contracts, public policy, and the practical alternatives available to each side. Institutions help decide whether a productive advance becomes a broadly shared one.

A business that invests in a useful system has a legitimate claim to a return. Someone took a risk, paid for implementation, and organized the work. Workers also have claims. Their experience may make the system usable; their efforts sustain the service; their knowledge may be embedded in the examples from which the organization learns. Recognizing these contributions does not require an identical share for everyone. It requires rejecting the fiction that only one party creates value.

A wage increase is one possible way to share the gain. Another is a reduction in unpredictable overtime. More stable schedules, better training, safer work, and fewer exhausting routines can also matter. The exact bargain will vary. A worker caring for a relative might value predictable time more than a small bonus. Another might prefer additional earnings. A serious process gives these preferences somewhere to enter the decision.

Public institutions have comparable choices. If a tax office processes forms more efficiently, citizens might receive quicker service, staff might gain time to resolve complex cases, and the government might reduce costs. The public deserves a clear account of the intended benefit. Efficiency should not become a word that excuses deteriorating access while celebrating an internal metric.

We should also ask whether the gain is as large as announced. The purchase price of a tool can be lower than the full cost of using it. Staff may spend time correcting mistakes, investigating uncertain outputs, or adapting information to incompatible systems. A fair assessment includes those costs. Otherwise, the organization may count time saved by one employee while overlooking work shifted onto another.

The burden can even move to customers. Imagine an automated service that reduces a company’s staffing costs but requires callers to repeat information, challenge wrong answers, and search for a human contact. The company’s productivity measure may improve while everyone else’s time is treated as free. A social assessment should count the whole arrangement. The person whose afternoon disappears into correcting an automated error has contributed a real cost, even if it never appears in the provider’s accounts.

None of this means protecting every existing business model from competition. A society can welcome new firms that provide a genuinely better service. It can also insist that people harmed by a transition deserve more than a lecture about adaptability. The case for innovation becomes stronger when the institutions promoting it accept responsibility for the foreseeable burdens of change.

The question of reward is ultimately a question of membership. If people are asked to accept disruption because the economy will become richer, they can reasonably ask how they will participate in that richer economy. A promise of aggregate prosperity becomes politically fragile when the route from national success to household security remains unexplained.

A voice before the system arrives

Workers often know details of a job that are invisible in a procurement document. They know which customer questions are easy to answer and which are easy to misunderstand. They know when a shortcut produces more work later. They may recognize that an apparent inefficiency is actually a precaution against a rare but serious problem. Excluding this knowledge can make a technological change worse even on its own managerial terms.

Consider an invented warehouse evaluating an AI scheduling tool. Managers want to reduce idle time. Staff explain that certain breaks in the schedule allow equipment checks, help with awkward loads, and recovery after physically demanding tasks. A system that treats every gap as waste could increase risk or fatigue. A well-designed process would investigate the workers’ account and incorporate what it learns into the objectives.

Worker participation does not mean that every individual receives a veto over every operational change. It means that people affected by a system have a credible opportunity to shape it, understand it, and challenge harmful outcomes. Representation can take different forms across workplaces. What matters is whether participation happens early enough to alter the plan and whether raising a concern carries a penalty.

A pilot designed with workers could examine whether the tool saves time, shifts burdens, affects quality, or changes the pace of the day. It could identify circumstances in which the system should be bypassed. It could distinguish a learning period from permanent performance expectations. This is a practical proposal: learning from those who perform the work can improve both the technology and the legitimacy of its use.

Training belongs in that process, but it is often asked to solve too much. Reskilling is useful when there is a realistic connection between the skill taught and worthwhile opportunities. It is less useful when a worker receives a generic course while suitable jobs remain distant, scarce, poorly paid, or incompatible with caring responsibilities. A certificate cannot move a household, create affordable childcare, or guarantee that an employer will hire its holder.

The time required to learn also has to come from somewhere. If an organization benefits from a new system, it should not automatically expect workers to bear the entire cost of adapting outside paid hours. Some may enjoy independent experimentation. Others may be exhausted, working several jobs, or supporting family members. Treating enthusiasm as the only legitimate response converts differences in resources into judgments about character.

Transition support should therefore be concrete. Paid learning time, access to the actual tools, realistic placement arrangements, and temporary income protection can be more meaningful than a broad promise to prepare people for the future. These measures have costs and should be evaluated honestly. So does leaving people to absorb every shock privately. A policy that ignores the second cost is not necessarily cheaper; it may simply place the bill elsewhere.

Organizations should also preserve routes through which inexperienced people become capable. If entry-level work changes, employers need to ask how newcomers will acquire judgment. It may be possible to design supervised practice using AI, simulations, feedback, and progressively more difficult responsibilities. That is a task for educational and organizational invention. Eliminating a familiar training route without creating another can leave a future shortage of the expertise everyone assumes will remain available.

The aim is not to freeze work. It is to make change something people can help shape. A worker who understands the purpose, contributes to the design, receives useful training, and shares in the benefits has stronger reasons to support a new system. These arrangements offer a stronger basis for trust because workers can examine the purpose, influence the changes, and see how the benefits are shared.

Worth after the job description

There is a final question that labor-market analysis cannot settle: what is a person worth when a machine can perform the task for which that person was paid? The answer should not depend on forecasting how many new occupations will appear. Even if every displaced worker eventually found another role, human dignity would still need a foundation more secure than temporary market demand.

A parent caring for a child, a neighbor helping someone recover, a retired person preserving a community’s memory, and a sick person receiving care do not become less human because their activities generate little income. Paid work can be a powerful source of independence, contribution, skill, and companionship. It is not the source from which a person’s right to respect is manufactured.

This distinction makes it possible to defend good employment without making employment sacred in every form. Some jobs injure people, exhaust them, or require hours of avoidable repetition. If technology removes part of that burden, we should be able to call it an improvement. The moral problem arises when freedom from a task is delivered as freedom from an income, a future, and a place in society.

Imagine a genuinely more productive economy that needs fewer hours of routine labor to provide essential goods and services. There would be reason to explore shorter working time, stronger public services, broader participation in ownership, or more generous protection during transitions. These are different political options with different costs and tradeoffs. No single one follows automatically from AI. What follows is the need to make an explicit decision about how abundance will support human lives.

A shorter week could offer more time for family, study, art, civic involvement, or rest. It could also fail if reduced hours mean inadequate income or if work intensifies until four days feel like six. A public benefit could improve security but be designed so poorly that access becomes humiliating. Shared ownership could broaden rewards while exposing people to risks they do not understand. Serious optimism examines the arrangement, not just the appealing label.

Meaning presents a related challenge. People who lose work may lose a valued community and a sense of usefulness as well as money. It would be arrogant to tell them that they should simply enjoy leisure. It would be equally narrow to assume that only a conventional job can provide purpose. Education, care, voluntary service, creative practice, and civic participation can offer real contribution when institutions give them room and respect.

A society that promises liberation from labor should therefore ask what people will be free to do, with what resources, and in whose company. Freedom becomes more substantial when a person can choose among worthwhile activities without constant fear of destitution. It becomes less substantial when they are told they are free while every practical route to belonging has disappeared.

These choices reveal the difference between respecting intelligence and worshipping productivity. More capable tools can help us build an economy with less waste, safer work, and greater room for human development. That is an ambitious and defensible goal. But no improvement in computational performance can decide that an unemployed person matters less, or that those who own the tools have earned the right to define everyone else’s worth.

The measure of success is therefore larger than output per hour. It includes the lives made possible by the hours saved, the security offered during change, and the standing retained by people whose work is transformed. The question is not only what our machines can produce. It is whether greater productive power helps more people live as members of a shared society, able to contribute, to choose, and to be valued beyond the latest demand for their labor.

Sources and notes