A few weeks ago I wrote about building my personal website in one evening. One person. Work that used to require a developer, a writer, a designer, and someone to coordinate between them. The output was the same. The team was not (Díaz, 2026).
I framed that piece around a question: when AI concentrates economic output in fewer hands, who benefits? I want to push that question further. Not toward individuals this time, but toward governments. Because the math that troubled me at the personal level turns out to be catastrophic at the national level, and almost no one is treating it as a fiscal problem yet.
Here is the problem in plain terms. Social systems, Social Security, Medicare, public pensions, are funded by taxing labor. Specifically, by taxing the wages of working-age people. That base is shrinking in two directions at once. Demographically, because fewer people are being born. Economically, because AI agents are absorbing the work those people used to do. The output remains. The payroll taxes don't.
The Demographic Half
South Korea's total fertility rate hit 0.72 in 2023, according to Statistics Korea (2024). That is the lowest ever recorded anywhere. Not just the lowest among developed nations. The lowest on the planet. A country needs a rate of 2.1 to keep its population stable without immigration. South Korea is producing births at one-third that rate. By 2100, its working-age population is projected to shrink by nearly 80 percent (Terry, 2024).
China's trajectory is slower but larger in absolute terms. The UN's 2024 World Population Prospects report projects that China's population will fall from 1.4 billion today to roughly 633 million by 2100 (United Nations, 2024). Its working-age population is projected to drop from 984 million in 2024 to 745 million in 2050. That is a loss of 239 million workers in 25 years (Visual Capitalist, 2025). A RAND Corporation analysis published in 2025 put China's labor force contraction at 28 percent by 2050 from its 2015 peak, with the old-age dependency ratio more than doubling over the same period (Pollard et al., 2025).
Japan recorded more than twice as many deaths as new births in 2023. Its fertility rate is 1.2 (Al Jazeera, 2024).
The United States is not in freefall by comparison, but the direction is the same. The Congressional Budget Office projected in early 2026 that the U.S. total fertility rate was 1.60 in 2024 and will decline to 1.53 by 2036. The ratio of working-age adults to retirees was 2.9-to-1 in 2024. CBO projects it will fall to 2.2-to-1 by 2054. In 1960, it was more than 5-to-1 (Congressional Budget Office, 2026).
That compression is not abstract. The Social Security Trustees reported in 2025 that Social Security's primary trust fund will be depleted by 2033. After that point, the program can pay approximately 77 percent of scheduled benefits from current payroll tax revenue alone (Social Security Administration, 2025). The shortfall is not a budget anomaly. It is a headcount problem.
The Economic Half
Demographic decline is a known problem. Governments have been modeling it for decades. What they have not modeled well is what happens when AI agents start absorbing the economic activity that human workers used to perform, at the same time populations are shrinking.
I am not talking about fears of full automation. I am talking about what is already happening. Salesforce eliminated 4,000 customer support roles in 2024 and attributed the reduction directly to productivity from AI agents (Fortune, 2025). The work those roles performed still gets done. Customer queries are answered, cases are resolved, issues are escalated. The economic output exists. The 4,000 W-2s do not.
A New York Times report found that AI was cited in the announcements of more than 50,000 layoffs in 2025, with many of those companies having no mature AI systems ready to replace the roles being cut (New York Times, 2026). The jobs went first. The tax revenue went with them.
The Brookings Institution published a framework in January 2026 specifically flagging this gap. As AI systems take on more autonomous economic roles, they projected that revenue from payroll taxes as a share of GDP will decline just as the need for retraining programs and social support is rising. They described this as a structural fiscal problem, not a cyclical one (Brookings Institution, 2026).
This is where the two forces meet. Fewer humans being born means a smaller future tax base from labor. AI agents absorbing current economic activity means a smaller present tax base from labor. Both are happening simultaneously. The social systems built on payroll revenue are caught in the middle.
The Legal Fiction We Already Built
I want to be precise about what I am claiming here and what I am not.
I am not arguing that AI agents are people. I am not arguing they deserve rights. I am arguing that we have solved a structurally similar problem before, and the solution was useful enough that we still use it 150 years later.
When industrial-era corporations became large enough to act as independent economic actors, the law had no clean way to address them. They could not own property, sign contracts, or be taxed as entities in their own right. They existed in a legal gap. The solution was corporate personhood, a deliberate legal fiction. Not a philosophical statement about the nature of corporations, but a governance tool. We gave corporations the legal standing necessary to participate in economic and fiscal systems. We did it because it was practical, not because we believed corporations were conscious.
The Yale Law Journal noted in 2024 that this fiction "has been granted to entirely fictional corporations without controversy" (Yale Law Journal, 2024). The EU Parliament explored a version of this concept under the label "electronic personhood" for AI systems, though it has not been adopted into law (Lexology, 2025).
The relevant precedent is not that corporations have rights. It is that legal personhood has historically been extended to non-human entities when governance needed it. The question is whether we need it again.
What a Framework Could Look Like
This is speculative. I want to be clear about that. But speculative framing grounded in real structure is how policy ideas become policy proposals, so it is worth sketching.
The core idea is not to tax AI systems as if they were workers. That would be both philosophically confused and practically unworkable. The idea is to define a category of economically active AI agent, an agent that autonomously performs tasks that generate measurable economic value, and to levy a tax on that activity, with the deploying organization as the responsible party.
The tax base would not be the agent's existence. It would be its economic output. Task completions, contracts executed, decisions made autonomously within defined parameters. This is not conceptually different from how we tax corporate activity: we do not tax a company for being incorporated, we tax the revenue and income it generates.
The deploying organization bears the obligation. Not the model provider. Not the infrastructure vendor. The entity that configured and deployed the agent to perform economically productive work. This preserves accountability in a place where it can actually be enforced, a registered legal entity with a tax ID.
The revenue would flow to the social systems most directly affected by labor displacement. Social Security, workforce retraining programs, and the public health infrastructure that supports aging populations. Not into a general fund where the connection disappears.
There are real objections. How do you define economic activity for an agent that is continuously running? How do you prevent deployment shifting to low-tax jurisdictions, the same problem that haunts corporate tax today? How do you draw the line between an AI agent and a piece of software that has always automated tasks? These are not small questions.
Xavier Oberson, a professor of tax law who has written specifically on this topic, put it plainly: taxing AI entities is a way to recoup revenue losses from worker displacement. Reuven Avi-Yonah at the University of Michigan framed it differently: taxation as a regulatory device, a lever you can adjust on an ongoing basis more easily than direct command-and-control regulation (McMahon, 2025). Both framings have merit. They are not mutually exclusive.
The harder objection is competitive. Governments that move first on AI taxation risk slowing adoption and losing ground to jurisdictions that do not. This is the same argument that has delayed meaningful corporate tax reform for two decades. It is also the argument that leaves social systems underfunded while the problem compounds.
The Framing Choice
Here is what I think matters most.
The current policy conversation treats demographic aging and AI displacement as separate problems. Aging is addressed through benefit reform: raise the retirement age, adjust the formula, find ways to make payroll revenue stretch further. AI displacement is addressed through labor policy: retraining programs, transition support, maybe a robot tax someday.
Neither addresses what happens when both forces hit the same tax base at the same time.
South Korea declared a "Population National Crisis" in June 2024 and created a ministry for it (Georgetown Journal of International Affairs, 2024). The United States Social Security system is running a $67 billion annual deficit as of 2024 and will exhaust its reserves in under a decade (Social Security Administration, 2025). China is looking at the loss of a quarter of its workforce by 2050 (Pollard et al., 2025).
Meanwhile, AI agents are processing insurance claims, answering support tickets, writing code, generating reports, executing trades. The work is happening. The taxpayer is not showing up.
The framing choice is whether governments treat this as a labor problem or a tax base problem. A labor problem leads to benefit cuts and retraining programs. A tax base problem leads to asking who, or what, should be contributing to systems that the whole economy depends on.
The demographic math is already in motion. The agent economy is already here. The taxman has not caught up.
References
Al Jazeera. (2024, February 28). Fears for future as South Korea's fertility rate drops again. https://www.aljazeera.com/news/2024/2/28/fears-for-future-as-south-koreas-fertility-rate-drops-again
Bipartisan Policy Center. (2025, November 19). 2025 Social Security trustees report explained. https://bipartisanpolicy.org/article/2025-social-security-trustees-report-explained/
Brookings Institution. (2026, January 8). The future of tax policy: A public finance framework for the age of AI. https://www.brookings.edu/articles/future-tax-policy-a-public-finance-framework-for-the-age-of-ai/
Congressional Budget Office. (2026). The demographic outlook: 2026 to 2056. https://www.cbo.gov/publication/61994
Díaz, S. (2026, March 4). One evening. One person. One website. Santiago Díaz. https://www.santiago-diaz.com/blog/one-evening-one-person-one-website
Fortune. (2025, September 2). Salesforce CEO Marc Benioff says his company has cut 4,000 customer service jobs as AI steps in. https://fortune.com/2025/09/02/salesforce-ceo-billionaire-marc-benioff-ai-agents-jobs-layoffs-customer-service-sales/
Georgetown Journal of International Affairs. (2024, September 24). The necessary paradigm shift for South Korea's ultra-low fertility. https://gjia.georgetown.edu/2024/09/24/the-necessary-paradigm-shift-for-south-koreas-ultra-low-fertility/
Lexology. (2025, September 5). Why future AI may deserve legal protection from human harm. https://www.lexology.com/library/detail.aspx?g=3cbb6ba2-1ca2-4ee7-89bb-e312551eca94
McMahon, K. (2025, November 4). Why we need to think about taxing AI. Transformer News. https://www.transformernews.ai/p/why-we-need-to-think-about-taxing
New York Times. (2026, February 1). Layoffs, AI, and the gap between announcement and reality. https://www.nytimes.com/2026/02/01/business/layoffs-ai-washing.html
Pollard, M. S., Bouey, J., Wang, A. X., & Pandey, R. (2025). Fertility decline in China and its national military, structural, and regime security. RAND Corporation. https://www.rand.org/pubs/research_reports/RRA3372-1.html
Social Security Administration. (2025). Trustees report summary. https://www.ssa.gov/oact/trsum/
Statistics Korea. (2024). Birth statistics 2023. Korean Statistical Information Service. https://kosis.kr
Terry, R. (2024, March 27). Lessons from South Korea's fertility freefall. The Terry Group. https://terrygroup.com/lessons-from-south-koreas-fertility-freefall/
United Nations Department of Economic and Social Affairs, Population Division. (2024). World population prospects 2024: Summary of results. United Nations. https://population.un.org/wpp/assets/Files/WPP2024_Summary-of-Results.pdf
Visual Capitalist. (2025, December 5). Charted: India vs. China working age populations (2024–2050). https://www.visualcapitalist.com/charted-india-vs-china-working-age-populations-2024-2050/
Yale Law Journal. (2024, April 22). The ethics and challenges of legal personhood for AI. https://yalelawjournal.org/forum/the-ethics-and-challenges-of-legal-personhood-for-ai