Labor Day and the Price of Disposability
AI did not invent the habit of treating workers as expendable. Fights over automation, water, and taxation reveal how much of that habit we are willing to carry into the future.
Labor Day was created to honor the people whose work produced American prosperity. It is an awkward holiday for an economy that has become increasingly sophisticated at obtaining work without accepting the obligations once attached to employment.
Labor economist Paul Osterman estimates that more than one in three American workers occupies what he calls a “disposable” job. His definition includes contractors, organizational freelancers and employees whose positions offer little security or opportunity for advancement. These are not necessarily short-term workers. Some perform essential work for years while remaining outside the durable relationships through which employers traditionally provided benefits, training and career prospects.
The arrangement has a familiar economic appeal. Employers gain flexibility, reduce costs and transfer some of the risks of uncertain demand onto workers. But the consequences extend beyond individual paychecks. Osterman’s research points to lower earnings and job satisfaction among several categories of disposable workers, as well as public costs when outsourcing undermines safety or service quality (Osterman, 2026).
AI is entering a labor market that has already learned how to make workers interchangeable. The question is not just about how many jobs automation will eliminate. It is whether the technology will intensify an older practice: obtaining productive capacity while shifting the associated obligations somewhere else.
The technology industry has its own history with that practice.
In 2000, Microsoft agreed to pay $97 million to settle litigation brought by long-term temporary workers excluded from employee benefit plans. In Vizcaino v. Microsoft, workers classified as independent contractors had performed work sufficiently integrated into the company’s operations that the courts found many were common-law employees. Their exclusion from certain benefits became the subject of years of litigation (120 F.3d 1006, 1997; 142 F. Supp. 2d 1299, 2001).
These workers became known as permatemps: permanent enough to perform the work, temporary enough to remain outside the protections of permanent employment. Microsoft subsequently imposed tenure limits and required breaks in service for temporary workers. The company disputed that those policies were connected to the lawsuit, but the arrangement illustrated how employment boundaries could be redesigned to manage legal and financial exposure rather than necessarily create a more durable relationship with the people doing the work.
That history gives Bill Gates’s renewed interest in taxing automation a significance beyond the usual debate about whether robots will take our jobs. Gates first proposed a robot tax in 2017, arguing that governments should consider taxing automation that replaces human workers to help finance the social costs of displacement. In August 2026, reporting on a new essay attributed to him a broader proposal involving taxes on robots and AI-token usage, as well as the idea of reserving certain kinds of work for humans.
The logic is that employers pay payroll taxes when they hire people, while investments in machinery receive different tax treatment. If automation allows companies to capture productivity gains while workers and governments absorb the costs of displacement, the resulting market incentives may not reflect the full social cost of substitution.
It would be easy to call Gates a hypocrite and stop there. The Microsoft settlement does not establish his personal motives, and no evidence shows that his current proposals represent a confession about the company’s earlier employment practices. But the institutional contradiction deserves attention. The technology industry helped demonstrate how productive work could be separated from the obligations of employment. Gates is now arguing that AI may extend that separation on a far larger scale.
That may represent learning from past mistakes. The meaningful test is whether the industry accepts obligations it can no longer engineer away.
The same question is emerging around AI’s physical infrastructure.
In June, a fabricated quotation attributed to Jeff Bezos circulated online, claiming he argued that water should be prioritized for AI over human needs. He had not. A fact-check by Snopes found no evidence that Bezos made the statement during his VivaTech appearance or elsewhere.
The quotation was false. The anxiety that made it believable was not.
AI infrastructure requires electricity and cooling. Data centers can consume water directly through cooling systems and indirectly through electricity generation, while semiconductor manufacturing adds further demands. Yet no single meaningful figure exists for the water an AI query consumes. A 2025 study found that water use per workload can vary by more than 10,000-fold depending on factors including cooling technology, server efficiency, climate, and the electricity grid.
Variability matters. A facility using reclaimed water in a region with abundant supply presents a different problem from one drawing freshwater in a water-stressed community. The relevant question is not whether AI uses water, but how much, where, under what conditions, and at whose expense.
The coming “water war” is therefore unlikely to resemble the cartoon implied by the Bezos hoax: a billionaire standing beside a reservoir choosing computers over thirsty children. It will be more ordinary, and harder to see. It will occur through siting decisions, utility contracts, tax incentives, cooling standards, and negotiations over who pays for new infrastructure.
A community may never be explicitly told that computation is more valuable than its water. It may just discover that the economic system has allocated the resource as though it were.
This is where the labor and water debates converge because both workers and water reveal the limits of an accounting system that records benefits where they accrue while allowing costs to surface elsewhere. Workers and water are not equivalent.
A robot tax is one possible response, but it is not a complete answer. Tokens measure computation, not displaced jobs. AI that helps a nurse document patient care is not economically equivalent to AI introduced principally to eliminate a customer-service department. A uniform usage tax could penalize beneficial applications while failing to distinguish augmentation from substitution.
Nor should automation be treated as inherently harmful. Productivity gains can reduce dangerous work, improve medicine, lower costs and create new forms of employment. The challenge is to distribute those gains without assuming that the people displaced by a transition can absorb its consequences.
A more durable response would combine appropriate taxation with portable benefits, wage insurance, retraining, stronger worker bargaining rights and mechanisms for sharing productivity gains. Environmental governance would require equally serious attention to local water conditions, transparent reporting and community authority over infrastructure decisions.
Gates’s reported proposal for Human Reserved work introduces another tension. Recognizing that some activities should remain human even when machines can perform them has value. Care, education, and civic judgment involve relationships and responsibilities that cannot be reduced to technical output. But the “reserve” metaphor risks implying that everything outside the protected boundary is available for replacement.
With technology, humans are always asking from a place of deficit instead of abundance. The more durable question is not “which jobs humans must be allowed to keep.” It is, “What obligations accompany the substitution of human labor at all?”
Labor Day’s original proposition was not that work should remain unchanged. Industrial capitalism was already transforming work when the holiday emerged. The proposition was that productivity creates obligations, and that (all) The People producing prosperity deserve a say in how its benefits are shared.
AI policy will fail if employment, environmental resources, and taxation remain separate conversations. They are becoming one conversation about distributing technological returns.
We do not need to choose between opposing AI and surrendering to it. Nor do we need fabricated villains declaring that machines deserve water more than people. We need institutions capable of asking a more difficult, nuanced question before markets answer it for us:
When technology produces more with less, who receives the return, who bears the risk, and who has the authority to decide?
That is the Labor Day question worth carrying into the next technological revolution.