For years, we have heard the same warning: artificial intelligence is going to destroy jobs. What no one anticipated clearly enough is which jobs would go first. Not executives. Not senior consultants with twenty years of experience and a network built over corporate dinners. The first to fall are those who still have none of that: juniors. Newcomers. The people still trying to get in.
That changes the usual story quite a bit.
The data worth reading slowly
A joint study by Harvard and Revelio Labs analyzed the résumés of 62 million people across 285,000 companies and found something uncomfortable: when a company adopts generative AI, junior employment falls by 7.7% over six quarters. Senior employment, by contrast, stays stable. Or grows.
It is not that AI is firing people en masse — the mechanism is quieter than that. Companies simply stop hiring at the bottom. They do not renew that junior analyst opening. They do not post that entry-level technical support position. And so, without drama or headlines, the door slowly closes.
Stanford Digital Economy Lab documented it from another angle: workers between the ages of 22 and 25 in occupations with high AI exposure have seen employment fall by 6% since 2022. Workers over 35, during the same period, grew between 6 and 9%. The World Economic Forum adds that entry-level vacancies in the United States fell by 35% in just 18 months. In the United Kingdom, job ads for junior developers dropped by nearly 45% in one year.
This is not a trend. It is a restructuring.
Business logic — and what it leaves out
The business logic is understandable, though that does not make it good. A junior employee, in most companies, does three things: prepares reports, gathers information, and executes well-defined tasks. Exactly the kind of work a language model can do in seconds and without asking for a raise. From the perspective of a CFO staring at a spreadsheet, the arithmetic is obvious.
What that arithmetic does not capture is what gets lost in the process. A junior employee is not valuable only because of the tasks they perform today. They are valuable because of the professional they could become in five years, if someone gives them the chance to make mistakes, learn, and grow within an organization. By eliminating that role, companies are not only cutting costs now: they are hollowing out their own future.
Mor Segal, from Five Sigma, called it directly a "generational gap." Who is going to be the senior professional of 2035 if in 2025 we hire no one?
There is something deeply short-sighted about this shift, and it is surprising that it has not sparked more debate.
The angle no one discusses: generational inequality
Entry-level jobs have historically been the first step of mobility for people who do not have networks or family names that open doors. If that step disappears, this is not just a labor-market problem. It is a problem of social structure.
More than half of Gen Z, according to the Young Policy Institute, fear being replaced in the job market by peers with stronger AI skills. That fear is not irrational. But it is also not exactly the right fear: the problem is not that another human will take the role by using ChatGPT better. The problem is that the role may not exist at all.
What do we do with all this?
Easy answers are everywhere: "learn AI," "build soft skills," "reinvent yourself." It is not that these are bad recommendations — they are necessary. But there is something condescending about placing the full burden of adaptation on the shoulders of the youngest and least powerful, while companies optimize margins and governments stand by watching.
The organizations responding well are not the ones that simply automated the bottom layer and called it a day. They are the ones that redesigned entry-level roles so they still make sense in an AI environment: supervising outputs, checking results, adding context, and asking the questions the model does not know to ask itself.
IBM tripled its hiring of recent graduates in 2026 and put them to work with real clients and AI tools. Cognizant integrated 25,000 recent graduates in 2025 into roles focused on analysis and oversight. It is possible. It requires investing in training and accepting that a junior employee will not produce at maximum capacity from day one — something that has always been true, but that the pressure for immediate efficiency tends to erase.
The blind spot in the dominant narrative
The dominant narrative about AI and employment has an important blind spot: it talks a lot about which jobs are going to disappear, and very little about who will bear the cost of that disappearance. The data point quite consistently to the same answer: for now, that cost is being paid by those least able to absorb it.
That is not inevitable. It is a choice — dispersed, uncoordinated, made company by company in spreadsheets — but a choice nonetheless. And like any choice, it can be made differently.