On a gray morning in Manhattan, commuters hurried past a sight that felt both familiar and unsettlingly new. Taped to a streetlight was a sheet of printer paper, the kind usually reserved for lost cats or neighborhood yard sales. Instead, it carried the face of a laid-off tech worker, a short plea for work and a QR code linking to his LinkedIn profile.

“I thought that would make me stand out,” said software engineer Glenn Kugelman, who had turned a lamppost into a last-ditch job board. In another era, a résumé like his would have been enough to guarantee multiple offers. Today, it is a symbol of a deeper shift: the quiet collapse of assumptions that have guided an entire generation’s career choices.

That same anxiety is echoing far from Manhattan. At UC Berkeley, one of the world’s most prestigious computer science programs, professor James O’Brien has been hearing from students who did everything right. Perfect 4.0 GPAs. Internships. Side projects. And now, no job offers.

“Tech degrees no longer guarantee a job,” O’Brien wrote in a widely shared post, contrasting the current drought with the recent past, when Berkeley CS graduates routinely fielded multiple lucrative offers before graduation. The pipeline from elite STEM degree to high-paying tech job has cracked, and no one seems sure how to fix it.

What is happening is not just another tech downturn. It is not the dot-com bust, nor the post-2008 hiring freeze, nor the pandemic whiplash. This time, the disruption is not primarily about overvaluation or macroeconomic shocks. It is about the arrival of a new kind of worker: machines that can do much of what white-collar professionals were told would always be “future-proof.”

In the span of a few months, unemployment in the IT sector jumped from under 4% to nearly 6%. Major tech companies have announced waves of layoffs while simultaneously pouring billions into artificial intelligence. Mark Zuckerberg has already predicted that AI will replace large numbers of mid-level engineers in the near future. The message is blunt: the very roles that defined the last tech boom are now on the chopping block.

For years, parents, teachers and policymakers pushed young people toward “safe” careers: software engineering, data science, medicine, corporate law. These were framed as rational, practical choices in a volatile world. But AI is now encroaching on the diagnostic side of medicine, the routine drafting and review work of law, and the bulk of coding tasks that once required teams of engineers. We have, in effect, overallocated our brightest students to fields that are suddenly vulnerable to automation.

This is not just about job loss. It is about a fundamental repricing of human labor. When machines can write code, summarize legal documents, analyze medical images and generate marketing campaigns in seconds, what is a human being worth in the marketplace?

One way to answer that question is to look at the kinds of work that remain stubbornly resistant to automation. These roles share four traits: they are valuable, scalable, ethical and what some researchers call “AGI-resistant” — meaning that even if we develop artificial general intelligence capable of performing almost any cognitive task, humans will still be preferred for these jobs.

That last criterion is the most elusive, because it is not purely technical. It is cultural and psychological. We already live in a world where machines outperform humans at many tasks, yet we deliberately choose the human version. Chess engines have long surpassed grandmasters, but human chess tournaments remain popular because they preserve the drama of error, risk and surprise. The Olympics bans most performance-enhancing technologies to protect the spectacle of unaugmented human ability. In both cases, we are not optimizing for performance. We are optimizing for meaning.

This preference for the human element shows up in less glamorous settings as well. In early childhood education, social work and community organizing, the value lies not in perfect information processing but in empathy, presence and trust. In restaurants and hotels, customers routinely pay more for attentive human service than for efficient automation. A robot can deliver a plate to your table; it cannot make you feel seen.

All of this points to a category of work that has been underestimated for decades: entertainment and human-centered service. The word “entertainment” conjures images of movie stars and musicians, but the category is far broader. It includes live performers, fitness instructors, tour guides, bartenders, therapists, coaches, teachers, event organizers and the armies of workers who create experiences rather than products.

What unites these roles are three psychological drivers that AI struggles to satisfy: boredom, loneliness and the desire for scarcity. People seek out experiences that feel alive, unrepeatable and shared with others. They want to belong to communities, to be recognized, to participate in stories that matter to them. An AI can simulate conversation, but it cannot truly share a room, a risk or a memory.

In a strange inversion of the advice given to millennials and Gen Z, the jobs that may prove most durable in an AI-saturated economy are not the most technical, but the most human. The bartender who remembers your name. The teacher who changes how you see yourself. The trainer who gets you through one more rep because you do not want to disappoint them. These are not side roles in a tech-driven economy; they may become its backbone.

That does not mean the transition will be smooth. In India, a country with 371 million young people, youth unemployment hovers in the double digits. In the United States, hundreds of thousands of high-earning professionals in software and related fields are at risk of displacement as companies adopt AI tools that can do in minutes what once took teams weeks.

Over the next three to five years, as AI systems grow more capable and capital shifts away from traditional software roles, the labor market could experience a shock. Large segments of the population may find their skills suddenly devalued. Governments have floated ideas like universal basic income, but political systems move slowly, and social safety nets are rarely built at the speed of technological change.

Any serious response will need to operate at scale. One proposal gaining traction among economists and policy thinkers is a modern version of the New Deal: a massive public works program focused not on roads and dams, but on the physical and digital infrastructure of the 21st century. That could mean upgrading power grids for renewable energy, building resilient transit systems, retrofitting buildings, expanding broadband and creating public spaces designed for community and culture.

Such a program would not only provide millions of jobs during the transition; it would also create the stage on which the next economy of human-centered work can flourish. Parks, performance venues, community centers, walkable neighborhoods and well-connected cities all make it easier for people to do the kind of work that AI cannot replace: caring, teaching, entertaining, hosting, healing and building relationships.

For students and early-career professionals, the implications are profound. Technical literacy will still matter, but as a complement rather than a guarantee. Skills in probability, statistics and data reasoning will pair powerfully with psychology, communication, design and community-building. The winners in the next era may be those who can understand systems and numbers, but also craft experiences that people actually want to show up for.

The uncomfortable truth is that no one has a complete roadmap. The Berkeley students with flawless transcripts and empty inboxes are right to feel misled by the old promise that a “practical” degree is a ticket to security. We cannot keep preparing young people for careers that may not exist by the time they graduate.

What we can do is adjust our lens. You are not simply competing with AI; you are being revalued in relation to it. The question is not “What can I do that a machine cannot?” but “Where will people still insist on another human, even when a machine is available?”

Those are the roles most likely to endure: the ones where humanity is not a bug to be optimized away, but the very feature people are paying for. The sooner we recognize that, the better chance we have to redesign education, policy and business around a future where machines do more work, but humans still do what matters.