Goldman Sachs Says AI Is Killing 16,000 Jobs a Month. Gen Z Is Paying the Price.

There is now a number.

For years, the debate about AI and jobs operated mostly in the territory of prediction. Studies estimated what could be automated. Researchers modelled what might happen. CEOs made claims about efficiency that nobody could verify. The honest answer to “how many jobs is AI actually destroying right now” was: we don’t really know.

This week, Goldman Sachs put a number on it.

According to research by Goldman Sachs economist Elsie Peng, AI has been erasing a net 16,000 US jobs per month over the past year. The breakdown is precise: AI substitution wiped out roughly 25,000 jobs per month, while augmentation — AI making existing workers more productive in ways that can support hiring — added back about 9,000. The difference is 16,000. Every month. For the past twelve months.

And the people absorbing almost all of it are under 30.

The Goldman research is more careful than most. It does not simply count layoffs and attribute them to AI. It uses a framework that combines standard AI exposure scores with a complementarity index developed by IMF economists, designed to separate two distinct forces operating simultaneously in the labour market.

The first force is substitution: AI handles a task, a human is no longer needed to do it, the job disappears. The second is augmentation: AI makes a human worker faster or more effective, potentially expanding what that person can accomplish and generating demand for more of that kind of work. The key insight is that both things are happening at once, in different parts of the economy, and the net effect is what matters.

Right now, substitution is winning.

The occupations scoring highest on substitution risk are exactly what you would expect from the broader research on AI and employment: insurance claims clerks, data entry workers, bill collectors, administrative support roles, customer service representatives. Routine, structured, text-based work that AI can handle without human judgment. These are also, not coincidentally, the jobs that a disproportionate share of young workers hold when they first enter the workforce.

This is the structural problem that the Goldman research makes unusually clear.

Gen Z workers are concentrated in the exact roles most exposed to AI substitution. Not by accident — by design. Entry-level jobs are, almost by definition, the jobs that involve the most structured, repeatable work. They are the jobs that exist precisely because they do not yet require the accumulated judgment and contextual expertise that takes years to develop. They are also, it turns out, the jobs that AI can handle most readily.

The unemployment gap between workers under 30 and workers aged 31 to 50 in high-substitution occupations has widened sharply relative to pre-pandemic averages. The wage gap has deteriorated too. Goldman’s regression analysis estimates that a one standard deviation increase in AI substitution exposure widens the entry-level to experienced wage gap by roughly 3.3 percentage points. Young workers in AI-exposed fields are not just losing jobs at higher rates. The jobs they can find pay less relative to their experienced colleagues than they used to.

I have written elsewhere on this site about what displacement looks like from the inside — about my sister who lost her translation work, about the legal training pipeline quietly collapsing, about Oracle sending 10,000 termination emails before breakfast. Those are the visible events. What Goldman’s research is tracking is something less visible: the slow narrowing of the door before it closes.

The people who are not getting hired do not show up in layoff announcements. They do not generate headlines. They are 22-year-olds submitting applications and hearing nothing, internships that did not convert, graduate schemes that had fifty spots this year instead of two hundred. They are the 14% drop in hiring of workers aged 22 to 25 into AI-exposed occupations that Anthropic’s researchers documented in March 2026. They are a generation running into a wall they can name now, even if nobody was naming it two years ago.

There is an augmentation story that deserves to be told alongside this one, because the Goldman research does not only contain bad news.

The jobs scoring highest on augmentation — where AI is expanding what workers can do rather than replacing them — are concentrated in roles that require genuine human judgment, creative problem-solving, and complex interpersonal work. Healthcare practitioners. Scientists. Engineers doing non-routine design work. These roles are not just surviving AI. They are, in some cases, growing because of it, as AI handles the structured elements and the humans focus on the parts that require being human.

The challenge for Gen Z is not that this augmentation story is false. It is that it describes work that requires years of foundational experience to reach. The engineers and scientists and practitioners who are thriving in an AI-augmented world got there through the entry-level years that are now disappearing for the people coming after them. The ladder is intact at the top. The bottom rungs are being removed.

This is the experience trap that nobody in the conversation about AI and jobs talks about honestly enough. You need experience to reach the roles AI cannot touch. You build experience by doing the entry-level work that teaches you things. The entry-level work is disappearing. The cycle that produced the experienced workers who are now thriving has been interrupted for the generation behind them.

What does this mean if you are under 30 and trying to figure out what to do?

The Goldman research does not offer career advice, and neither, honestly, do I. But the data points toward a few things that are worth taking seriously.

The roles with the highest augmentation scores — the ones where AI is making humans more valuable rather than replacing them — share a common feature: they require the kind of judgment, creativity, and interpersonal skill that takes time to develop and cannot be shortcut by a language model. Anything you can do to develop those capabilities now, before they become the only differentiator that matters, is time well spent.

The roles with the highest substitution scores are not a list of jobs to avoid at all costs. Some of them are still hiring. Some of them will be hiring for years. But they are hiring into a contracting market, and the people entering those roles now are building skills that may have a shorter shelf life than the career they are planning around them.

The most useful thing the Goldman research does is give the feeling that many young workers already have — that something has changed, that the job market is harder than it should be given their qualifications, that the rules seem to have shifted without anyone announcing it — a data-backed foundation. The feeling is correct. The rules did shift. 16,000 jobs a month, net, for the past year, concentrated among people who had not yet had the chance to become irreplaceable.

That is not a reason for despair. But it is a reason to take the question seriously, and to build accordingly.