If AI Can Do Every Job, Why Will It Need Us for the New Ones?

I need someone to explain this to me. If artificial intelligence is such a powerful technology that it will eventually perform most of the work humans do today, why will it need us to perform all the mysterious new jobs it supposedly creates tomorrow? If AI can replace accountants, programmers, designers, lawyers, writers, customer-service agents, analysts and eventually much of physical labour, what exactly prevents it from doing the new jobs too?

For years, the loudest voices in technology told us that AI was going to transform everything. It would become smarter than us, automate enormous amounts of human labour and create wealth on a scale we had never seen before. That promise helped justify massive valuations, unprecedented investment and a frantic race to build bigger models, larger data centres and more powerful systems. Elon Musk has gone so far as to predict a future in which “no job is needed” because AI will be able to do everything.

Now that people are beginning to ask whether replacing human labour is actually a desirable goal, the story is changing. Suddenly, we are told that AI will create millions of new jobs—better jobs, more interesting jobs and jobs we cannot even imagine yet. Whenever someone asks what those jobs will be, the answer is usually that no one could have imagined app developers or social-media managers before the internet. We are being asked to trust that the market will invent something.

Maybe it will, but that still doesn’t answer the question. If AI keeps improving as quickly as its builders claim, why would those future jobs remain human jobs for very long?

The Story Keeps Changing

The argument usually begins with history. Previous technologies destroyed some jobs and created others, so AI will presumably do the same. The automobile displaced parts of the horse economy but created mechanics, assembly-line workers, driving instructors, highway engineers and gas-station attendants. Computers eliminated some forms of clerical work but created software developers, IT departments, digital designers and thousands of new occupations.

That history matters, but it does not guarantee the same outcome this time. Earlier machines generally replaced a particular form of labour. They lifted heavier objects, moved faster, calculated more accurately or completed repetitive physical tasks. Humans moved into work that required language, judgment, creativity, communication and the ability to learn something new. Those were our protected territories.

AI is being built specifically to enter those territories. It can write, code, analyze, persuade, design, plan and increasingly operate other machines. More importantly, it can learn the structure of a new task and then reproduce it at scale. If we create a new occupation because AI has automated an old one, that occupation becomes another collection of tasks that can be observed, measured and fed back into the same system.

There are reasonable explanations for why humans might remain involved. Customers may prefer human interaction. Institutions may require human accountability. Some work depends on physical presence, trust, social context or judgment that is difficult to automate. Even OpenAI’s own work-transition framework identifies human interaction, institutional rules, physical presence and accountability as reasons jobs may survive even when AI can perform many of their tasks.

Those are valid reasons for humans to remain in certain kinds of work, but they are not evidence that enough new jobs will appear to replace the ones that disappear. They are reasons society might choose to preserve human participation, not proof that the technology will naturally produce a better labour market. That distinction is everything.

AI Is Not a Labour Policy

There is a strange sleight of hand in the way this conversation is presented. The technical capability is treated as inevitable, while the social outcome is treated as automatically positive. AI will become more powerful because progress cannot be stopped, but somehow the benefits will naturally distribute themselves through new jobs, higher wages and greater opportunity.

Technology does not decide who receives its benefits. Companies, governments, workers, laws and markets decide that. The same AI system can help one person become more capable, allow another company to eliminate half a department and give a third company enough leverage to dominate an entire industry. Calling all three outcomes “productivity” doesn’t make their human consequences the same.

The International Labour Organization estimates that one in four workers worldwide is employed in an occupation with some degree of generative-AI exposure. Its research suggests that transformation is currently more likely than complete replacement for many jobs, but it also stresses that the outcome depends on how the transition is managed. That last part tends to disappear from the sales pitch.

AI does not arrive with a built-in social contract. It does not guarantee that productivity gains will be shared with workers, that shorter working hours will replace layoffs or that people whose jobs disappear will own any part of the systems that replaced them. Those are political and economic decisions. If we don’t make them deliberately, the default outcome will be determined by whoever owns the technology.

The honest answer to the central question is that AI may create new jobs, but there is no economic law guaranteeing that those jobs will be numerous, durable or better. If the technology continues improving, many of those new jobs could be automated too. The idea that people will simply keep moving into an endless supply of uniquely human occupations assumes that AI will stop advancing at exactly the point where the labour market needs it to stop. Nothing about the way these companies describe their ambitions suggests that is the plan.

We Already Ran This Experiment With Social Media

We have seen a version of this movie before. Technology companies built social-media platforms they did not fully understand, deployed them across the global population and treated billions of people—including children—as participants in an uncontrolled experiment. The products were optimized for growth and engagement while governments, schools, parents and researchers struggled to understand what they were doing to attention, mental health, public trust and childhood itself.

The consequences did not arrive all at once. They accumulated slowly enough for the platforms to become essential before society fully recognized the scale of the problem. By the time meaningful questions were being asked, the companies were among the most powerful institutions on the planet and their products had become embedded in politics, business, culture and everyday life.

On August 26, 2026, Meta agreed to pay up to $17.1 billion in a settlement with 47 American states over allegations that Facebook and Instagram were designed in ways that harmed children. The agreement also requires significant child-safety changes, and it ended a federal trial before Mark Zuckerberg’s expected testimony. The settlement does not resolve every debate about social media or establish that every allegation was true, but it is an extraordinary measure of how serious the consequences became.

We cannot repeat the same sequence with AI: deploy first, deny the problems, fight regulation, settle the lawsuits and make safety changes after the technology has reshaped society. AI is more powerful and more broadly applicable than social media. It is moving into education, medicine, government, employment, culture, warfare, intimacy and identity at the same time. The consequences will not be confined to what we see in a feed.

I am not particularly interested in whether the people running these companies consider themselves good people. That isn’t the point. Their job is to protect their companies, expand their market position and increase their value. A system that depends on executives voluntarily sacrificing competitive advantage for the long-term public interest is not a system. It is a wish.

We Need to Decide Before the Decision Is Made for Us

I build AI systems through Reimagine AI, and I believe this technology can be an incredible tool. I have seen it expand creativity, make difficult ideas accessible and help people accomplish things that would previously have been impossible. This is not an argument against AI. It is an argument against pretending that the way we deploy AI is neutral or inevitable.

We need meaningful public thought about what should be automated, what should remain human and who should receive the value these systems create. We need labour policy, transparency, enforceable safety standards, protections for human work, clear rules around data and real accountability for companies deploying systems at population scale. Most importantly, those conversations have to happen while the technology is being built, not years later when the damage has become too expensive and embedded to reverse.

We should also stop accepting vague promises about future jobs as an answer to present-day disruption. What jobs are likely to be created? How many people will they employ? How long will those jobs remain beyond AI’s capabilities? Who will own the new productivity? What happens to people during the transition, and who is responsible if the transition never ends?

There may always be work humans choose to do because we value human presence, accountability, creativity and connection. But if those things are going to remain part of our economy, we will have to protect and value them deliberately. The market will not automatically preserve human work when replacing it is cheaper.

The question is not simply whether AI can do our jobs. The question is what kind of society we want to build when it can. If we wait for the heads of technology companies to answer that for us, the answer will inevitably serve their technology, their companies and their shareholders first.