I have said it once, and I will say it a thousand times: giving people hope that employment will remain the primary source of income in the Age of Artificial Intelligence is unwise.
Human intelligence has now been digitized. The consequence is that businesses will naturally gravitate toward what they perceive to be the best available intelligence. Traditionally, every business had to employ different individuals to perform the same type of work because human beings cannot be in two or more places at the same time. This biological limitation helped create millions of jobs across the economy.
Artificial intelligence, however, is software. The same AI agent can simultaneously serve thousands—or even millions—of geographically dispersed businesses. The same winner-takes-all dynamics that characterize software markets are likely to emerge for digitized employees in the form of AI agents. Businesses will not settle for less when they can access the best available intelligence at a reasonable price.
There will, of course, continue to be jobs for the people developing the world’s leading AI systems and for those building solutions that automate specific tasks—or entire occupations. However, these same AI companies will themselves make extensive use of AI, meaning they can operate with relatively small workforces. As a result, the number of new jobs they create is unlikely to offset the number of jobs their technologies eliminate. Their customers, mostly businesses, will also operate with a relatively small workforce.
It may also be true that, for many years, AI will automate individual tasks rather than entire occupations. However, betting against the eventual automation of whole jobs is a risky strategy when the companies developing these technologies openly state that this is their objective. As a worker, you may believe your profession can never be fully automated, but AI companies are investing billions of dollars to prove otherwise.
The idea that humans will always be better than AI at supervising AI agents is already being challenged. Models such as Claude Mythos have demonstrated exceptional ability in identifying software security vulnerabilities—so much so that they are being used to review the work of expert human software engineers for potential flaws. As AI systems become more capable at evaluating AI-generated output, relying on humans to review every AI-produced result increasingly becomes a productivity bottleneck. AI can generate work far faster than a human can review it, making human oversight itself a limiting factor as AI capabilities continue to advance.
Companies such as Mercor are hiring experienced professionals to train AI models to perform work at the standard of seasoned experts. The goal is to capture human expertise, encode it into AI systems, and enable those systems to deliver professional-quality work at scale.
AI companies like OpenAI, Anthropic, Google, Microsoft, Meta, and Amazon are literally embedding their engineers—often called forward-deployed engineers—within their corporate customers’ offices to help them automate their business processes, including, where possible, entire job functions. If AI companies can automate an entire job rather than just a subset of its tasks, they create significantly more value for their customers and, in turn, generate greater revenue for themselves. The financial incentive to automate entire jobs, not just individual tasks, is therefore built directly into their business model.
How can we, with a clear conscience, tell people they will continue to find employment when they will increasingly have to compete against digital versions of the world’s leading experts? People need protection from financial hardship, not false hope that adequate employment opportunities will remain available in the Age of Artificial Intelligence. We need scalable solutions that ensure everyone has, at the very least, enough food to eat and a decent place to live—even without employment. These are flesh-and-blood human beings whose lives and well-being are at stake, not merely economic statistics.
Saying that, historically, whenever a new technology emerged some people lost their jobs but eventually found new ones is not reassuring. AI is not simply another new technology. What if this time people do not find new employment? What is the plan then? Should humanity bear the consequences of predictions that turn out to be wrong? If advocates of Universal Basic Income are wrong and employment opportunities remain abundant, the consequences are relatively limited. People would continue working while the economy benefits from a stronger social safety net. However, if those who dismiss Universal Basic Income are wrong and AI does lead to widespread unemployment, the consequences could be severe. From a risk-management perspective, advocating for a Universal Basic Income is the safer course of action.
Advising people to learn how to use AI will only take them so far because the AI agents they will ultimately compete against for employment will also know how to use AI and other software through capabilities such as tool use and computer use.
Encouraging people to use AI to start a business is sound advice. However, many digital businesses are likely to be subject to winner-takes-all dynamics, where only a small number of firms dominate a market while the majority are relegated to the long tail, generating little revenue. By contrast, brick-and-mortar AI franchise businesses are more likely to create broad-based entrepreneurial opportunities. These would be packaged, turnkey businesses operated primarily by robots and AI agents, while human franchise owners customize the customer experience and add their own unique touch.
As people accumulate Universal Basic Income payments, they could save and invest those funds to purchase AI-powered brick-and-mortar franchise businesses, allowing them to build wealth and participate in the AI economy as owners rather than relying solely on Universal Basic Income.
For those who want something more than a Universal Basic Income—those who value the dignity and sense of purpose that comes from productive work—investing in a brick-and-mortar AI franchise business is likely to be one of the best options. Franchises significantly reduce the high failure rates associated with starting a business because they are built on proven business models and come with ongoing operational support from the franchisor.
Those who are content living on a Universal Basic Income would remain financially protected. There is no need to sacrifice the security provided by Universal Basic Income simply to accommodate those who wish to continue working. In Aaron Kalikawe’s Policy for the Age of Artificial Intelligence, both aspirations can coexist: people can enjoy the financial security of UBI while those who seek additional purpose, income, or entrepreneurship remain free to pursue it.
The Solution by Aaron Kalikawe
For policymakers, the writing is on the wall: PLEASE HEED THE WARNING. You need to begin transitioning your economy from dependence on employment income to dependence on Universal Basic Income. The longer you wait, the more chaotic and painful that transition is likely to become. During the transition, governments will need to financially support those who become unemployed while UBI payments are still insufficient to meet their basic needs. The longer the transition is delayed, the greater the number of unemployed people who will require support.
The earthquake has already occurred beneath the ocean—the rise of artificial intelligence. It has triggered a tsunami that could lead to widespread job displacement. The waves have not yet reached the shoreline, and that gives policymakers valuable time to organize an orderly evacuation to higher ground by transforming their economies while most people are still employed. If you wait until the tsunami makes landfall before taking action, it may be too late. The transition to a UBI-based economy will not happen overnight, and governments may no longer have sufficient resources to support a rapidly growing unemployed population during the transition.
Professor Chad Jones discusses the weak links that could slow the high rates of economic growth expected from AI adoption. The danger of only partially automating the economy is that it may produce only modest productivity gains while simultaneously creating political unrest as the economic divide widens between those whose jobs remain unautomated and those whose jobs have been automated.
Weak links that slow economic growth occur wherever humans remain in the loop, particularly when performing tasks that can already be reliably automated by AI. The failure to automate these tasks may stem from a desire to preserve human employment or simply from a lack of awareness that the work can already be automated. In practice, this lack of awareness is one of the biggest reasons businesses fail to automate tasks that are already automatable.
This is one of the reasons my AI economic proposal advocates building entirely new AI-first Cities governed by AI companies with levels of autonomy comparable to Hong Kong and Macau, and contractually obligated to continuously develop AI systems that automate not only commercial activities but governance itself. Building new cities allows businesses to adopt an economic system that requires the automation of every task that can be reliably automated. The governing AI companies would build and maintain the infrastructure that enables businesses to automate their operations easily, including through third-party AI tools.
Humans are often the bottleneck in adopting new AI capabilities because workers require continual training whenever new technologies become available. By contrast, updating an AI model to use a newly released capability the next time it performs a task is relatively straightforward. Even the human brain has limits in its ability to remember every AI capability available for accomplishing a particular task. As AI models rapidly gain new capabilities—and as AI-assisted “vibe coding” accelerates the pace at which those capabilities are developed—it becomes increasingly impractical to expect human workers to remain continuously trained. Humans therefore become the bottleneck: the weak link that prevents AI adoption from keeping pace with technological progress. The result is that increasingly powerful AI models remain significantly underutilized because the people using them cannot keep up with their expanding capabilities.
The promise of AI-driven abundance will only be realized by maximizing automation throughout the entire economy. I do not believe conventional cities have the political will to pursue that level of automation without first protecting workers from the resulting job losses. That is why I propose building new AI-first Cities, where the economy has never depended on employment income and is designed from the outset to be sustained by Universal Basic Income.