Since the arrival of large language models into everyday working life, a quiet reckoning has begun in the offices and hiring halls of the modern economy — one that touches the oldest of human questions: what is labor worth, and who gets to perform it? Young workers entering finance, software, and creative fields are finding fewer doors open than their predecessors did, while the companies that once employed them speak openly of replacing headcount with autonomous systems. The story is neither catastrophe nor vindication, but something more characteristically human — uneven, contested, and stil
AI's Job Impact Emerges: Early Data Shows Mixed Displacement Across Sectors
Young workers aged 22-25 in AI-exposed sectors experiencing measurable employment displacement, with losses reaching 12.8% in finance, software, and creative industries.
Flat is the new up—the workforce stays level as AI agents multiply.
Mark
So we have data showing young workers losing jobs in software and finance, but companies are also rationing their AI use because it's too expensive. How do those two things fit together?
Mimi
They don't fit neatly, which is part of what makes this moment strange. The capability is there—these systems can genuinely do work that used to require humans. But capability and deployment are different things. A company might have a tool that can replace a worker, but if running that tool costs more than paying the worker, the math breaks down.
Mark
So the job losses we're seeing—the 12.8 percent drop for young workers in finance and software—that's happening even though companies are hitting the brakes on AI spending?
Mimi
Yes. And that's the unsettling part. The displacement is already happening at the margins, in hiring decisions, in how many entry-level positions get posted. Companies don't need to fully automate to reduce headcount. They just need to hire fewer people. They can do that while still figuring out the cost question.
Mark
What changes if cheaper Chinese AI models become the standard?
Mimi
Everything potentially. If the per-task cost drops significantly, the economic barrier to automation falls away. Right now, expensive AI might only make sense for high-value work. Cheap AI could make sense for almost anything. That's when you'd expect to see broader displacement.
Mark
But we don't know if that will happen?
Mimi
Not yet. We're watching it unfold in real time. The data we have is from the last few years—the early stage. The economists warning about this are essentially saying: we can see the capability, we can see some early job losses, and we don't know where the ceiling is. Act now, they're saying, because we might not get another chance to shape how this goes.
Il Polso
- Young workers aged 22–25 in AI-exposed sectors have seen employment drop by as much as 12.8% since ChatGPT's widespread adoption, with the losses sharpest in finance, software, and creative industries.
- AI systems have crossed a threshold — moving from handling seconds of skilled work to hours of it — and the next generation may begin developing itself, compressing timelines that once felt safely distant.
- The cost of deploying advanced AI has grown so steep that companies are rationing access to their own tools, revealing an economic ceiling that may slow automation's reach more than any policy or protest.
- Cheaper AI models originating from Chinese research are now entering Western markets freely, threatening to collapse the cost barrier and reopen the automation calculus for businesses that had paused.
- Nobel Prize-winning economists are urging governments to intervene before the shift becomes irreversible, while other researchers argue macroeconomic forces — not AI — explain the employment patterns so far.
Since the arrival of large language models into everyday working life, a quiet reckoning has begun in the offices and hiring halls of the modern economy — one that touches the oldest of human questions: what is labor worth, and who gets to perform it? Young workers entering finance, software, and creative fields are finding fewer doors open than their predecessors did, while the companies that once employed them speak openly of replacing headcount with autonomous systems. The story is neither catastrophe nor vindication, but something more characteristically human — uneven, contested, and still being written.
The executives steering the world's largest technology companies have settled on a blunt arithmetic: if artificial intelligence can perform work that humans currently do, the incentive to pay humans diminishes. The phrase circulating among investors — "flat is the new up" — captures the logic neatly: headcount holds steady or falls while output climbs, the gap filled by autonomous AI agents. Nobel Prize-winning economists have begun urging governments to act before this shift hollows out career paths rather than lifting living standards. The early data, however, tells a story that is messier than either alarm or reassurance.
The capability of these systems has grown with unsettling speed. Three years ago, large language models could reliably handle only tasks measured in seconds. Today they routinely absorb an hour of skilled work — identifying and rewriting flawed code in cryptocurrency contracts, performing entry-level financial analysis, drafting early-stage legal documents. Within a year, some systems may begin improving themselves. The acceleration is real.
Stanford University researchers tracking four years of American wage and employment data found that workers aged 22 to 25 have absorbed a 2.7% employment hit since ChatGPT became widespread — a figure that rises to 12.8% in finance, software, and creative industries. Some economists attribute the pattern to interest rate increases rather than AI, but the numbers remain. In the United Kingdom, job postings in highly exposed sectors like telemarketing and legal services declined sharply during a period when rates were stable and before new payroll taxes took effect — timing that points toward something structural rather than cyclical.
Yet the story pivots when you examine the actual cost of deployment. AI usage is measured in tokens — small text fragments, roughly three-quarters of an English word each — and consumption has exploded in 2026, driven by autonomous agents performing tasks without human oversight. The bills have grown so large that many firms are now rationing access to their most powerful models. Virtual workers, it turns out, can cost more than human ones.
Entering this equation now are cheaper AI models derived from Chinese research, released freely into global markets and already being adopted by Western companies. If capable automation becomes inexpensive, the economic brake that has slowed displacement may release. Whether AI ultimately hollows out labor markets or expands them may depend less on what the technology can do than on what it costs to do it — and that answer is still forming.
The executives running the world's largest technology companies are making a straightforward calculation: artificial intelligence can do work that humans currently do, and if it can, why pay for the humans? The pitch to investors has become almost a mantra—"flat is the new up," meaning a company's headcount stays level or shrinks even as output grows, replaced by what the industry calls "AI Agents," autonomous systems trained to handle specific tasks, some of them surprisingly sophisticated. Nobel Prize-winning economists have begun sounding alarms, urging governments to act now to ensure that this technological shift lifts living standards rather than hollowing out entire career paths. Yet the picture emerging from early data is neither a simple story of mass displacement nor one of seamless transition. It is messier, more uneven, and still unfolding.
The capability curve tells part of the story. Three years ago, the large language models that power most AI systems could reliably handle only the kind of work a human might complete in seconds or minutes. Today, they routinely tackle tasks that would consume an hour of skilled labor. The latest generation can identify bugs in cryptocurrency contracts, then redesign and optimize the code itself—work that would take a programmer several hours. Within the next year or so, these systems may begin developing themselves, at least in theory. The same pattern is appearing earlier in other fields: financial analysis, entry-level legal work, early-stage creative jobs. The capability is real and accelerating.
But what does capability mean for actual employment? The clearest evidence comes from the United States, where researchers have tracked four years of wage and job data across occupations most and least vulnerable to AI disruption. Stanford University's analysis found that workers aged 22 to 25 have experienced a 2.7 percent hit to employment since ChatGPT became widespread. In sectors most exposed to AI—finance, software development, creative industries—that figure jumps to 12.8 percent. Not every economist accepts this interpretation; some argue that interest rate increases and other macroeconomic factors better explain the pattern. But the data point stands: young workers entering these fields are finding fewer positions available.
Online job postings tell a similar story, though again with complications. The OECD measured the gap between job listings in highly exposed sectors like telemarketing and legal services versus less exposed fields like construction and food preparation. The United Kingdom showed a notably sharp decline in postings within exposed sectors at a time when interest rates were stable or falling, and before the government's National Insurance increase took effect. The timing suggests something other than general economic weakness. The UK's economy leans heavily on services—the very sector most vulnerable to AI displacement.
Yet the story takes a turn when you examine how much AI companies are actually deploying these systems. Usage is measured in "tokens," small chunks of text that AI systems process. One token roughly equals three-quarters of an English word. In 2026, token consumption has exploded, driven largely by what the industry calls "agentic use"—autonomous agents performing tasks without human intervention. Some companies have burned through trillions, even quadrillions of tokens in recent months. The bills became so staggering that many firms have begun rationing access to their most advanced models. This constraint matters. It suggests there may be economic limits to how much work can actually be automated. Virtual workers, it turns out, can be more expensive than human ones, depending on the task.
A complicating factor is now entering the equation: cheaper AI models derived from Chinese technology are being released freely into global markets. Western companies are beginning to adopt these alternatives, which could fundamentally alter the economics of automation. If a task can be performed by a less expensive system, the calculus changes. The question of whether AI will displace workers or augment them—whether it will hollow out job markets or create new ones—may ultimately hinge not on what the technology can do, but on what it costs to do it. For now, the answer remains uncertain, but the trends are becoming visible.
Citazioni salienti
The world must act now to ensure AI leads to rising living standards and not large-scale job displacement— Nobel Prize-winning economists
Virtual workers might be more expensive than human ones, depending on the task— Analysis of AI deployment costs
Vuoi la storia completa?
Leggi l'originale su BBC News ↗
Technology