The Entry-Level Vanishing Act: How AI Is Closing the Door on Young Workers

The most significant labour market shift of the AI era is not happening to people who already have jobs. It is happening to people who cannot get their first one.

Workers aged 22 to 25 in the most AI-exposed occupations have seen a 13% relative employment decline since late 2022. Not through redundancies. Not through layoffs. Through a quiet, systematic collapse in new hiring. According to research from the Dallas Federal Reserve, the door is not being slammed on people already in the building. It is simply not being opened for those trying to enter.

This is the story that the breathless AI productivity discourse keeps missing. While senior professionals enjoy AI-augmented workflows and rising wages, an entire generation is watching the bottom rungs of the career ladder dissolve beneath them.

The codifiable knowledge trap

The mechanism is surprisingly simple once you see it.

The Dallas Fed's research draws a distinction between two types of knowledge. Codifiable knowledge is the textbook stuff: rule-following, routine processes, the kind of structured tasks you can write down in a manual. Tacit knowledge is the judgement built from years of experience, the intuition that tells you when the process should be broken, the understanding of context that no documentation captures.

Entry-level white-collar work has always been dominated by codifiable tasks. Summarise this document. Draft that email. Process these records. Check this spreadsheet against that database. It is precisely this kind of work that AI now performs competently and cheaply.

Experienced workers, meanwhile, hold tacit knowledge that AI cannot yet replicate. For them, AI is a productivity amplifier. For junior workers, it is a replacement.

The result is a bifurcated labour market where the same technology that makes senior professionals more valuable makes junior professionals less necessary.

The numbers are stark

US entry-level job postings fell 35% between January 2023 and June 2025. In software engineering, the decline was 67% over 2023 to 2024 alone. The Stanford Digital Economy Lab found that employment for software developers aged 22 to 25 dropped nearly 20% from its late-2022 peak by July 2025. Their paper was titled "Canaries in the Coal Mine?" and the question mark felt generous.

The UK is no better. Tech graduate roles fell 46% in 2024, with a further 53% decline projected by 2026.

And here is what confirms this is structural rather than cyclical: wages in the top decile of AI-exposed industries grew 8.5% over the same period. In computer systems design specifically, wages rose 16.7% against a national average of 7.5%. Rising wages and falling employment in the same sector is a classic signal of structural change. The work is not disappearing. The people doing it are changing.

The experience premium tells the story

The Dallas Fed identified a useful predictor for which occupations are most affected: the experience premium, meaning the wage gap between entry-level and senior workers.

The median sits around 40%, but the range is enormous. Fast-food cooks have an experience premium under 10%. Lawyers and credit analysts sit above 100%. The pattern holds: the higher the experience premium, the stronger the AI-driven wage growth for seniors, and the steeper the decline in junior hiring.

This makes intuitive sense. Where experience matters a lot, AI amplifies the experienced and eliminates the need for juniors to do the groundwork. Where experience matters little, there is less to amplify, and less incentive to replace humans at all.

The exceptions prove the rule

Not every sector is following this pattern. Healthcare entry-level postings rose 13 percentage points. Skilled trades remain stable. Government roles have held steady.

The common factor is revealing: physical presence, unpredictable environments, and irreducible human interaction. You cannot automate a physiotherapist's hands or a plumber's assessment of a leaking pipe. You cannot replace the social worker who needs to read the room.

This is useful information for young people making career decisions right now. But it is cold comfort for the generation told to study computer science and learn to code.

The training paradox

Here is where this gets truly concerning.

The routine tasks that AI is automating away were never just busywork. They were training grounds. The Journal of Accountancy described this in March 2026 as a "training paradox" in professional services. The grunt work that junior accountants, lawyers, and analysts cut their teeth on was how they developed the tacit knowledge that eventually made them senior professionals.

Erik Brynjolfsson put it bluntly: "If you don't hire junior workers, you're not going to get senior ones."

The American Enterprise Institute frames the long-term risk as a collapse of the "skills commons," the shared base of practical experience that sustains an entire profession. If firms stop investing in junior talent because AI handles the entry-level tasks, the pipeline of experienced professionals dries up within a decade.

This is not a theoretical concern. It is a slow-motion crisis playing out across knowledge work right now.

The tax code makes it worse

There is a structural incentive problem that nobody in government seems eager to address.

According to the Brookings Hamilton Project, the effective tax rate on labour sits between 25.5% and 33.5%. The effective tax rate on automation capital is roughly 5%. That is a five to six times cost advantage for replacing a worker with technology.

The tax system was not designed to accelerate this substitution, but that is exactly what it does. Every employer making a rational cost calculation faces a system that actively penalises hiring humans and rewards deploying software. Until that imbalance is addressed, policy interventions like retraining programmes are pushing against a financial headwind.

What actually needs to happen

The honest answer is that there is no simple fix. But there are responses that would help.

Restructure entry-level roles around tacit knowledge acquisition. If the codifiable tasks are gone, design junior positions around the things AI cannot teach: client relationships, cross-functional judgement, ethical reasoning, stakeholder management. This requires employers to rethink what "entry-level" means, not just trim headcount.

Reform the tax imbalance. A technology that replaces workers should not enjoy a five-fold tax advantage over the workers it replaces. This is not anti-technology. It is basic market correction.

Invest in apprenticeship and mentorship models. The trades have always understood that you learn by doing alongside someone experienced. Knowledge work needs to rediscover this principle. AI can be part of this, as a training tool rather than a training replacement, but only if firms commit to actually hiring the people who need training.

Take the long view on professional pipelines. The quarterly cost savings from not hiring graduates look attractive now. The talent drought in five to ten years will not.

The door that never opens

GPT-5.4 launched on 5 March 2026, scoring 83% on the GDPval benchmark across 44 knowledge-work occupations. That is a 12.1 percentage-point jump from the previous version. The capabilities curve is steepening, and the category of "codifiable knowledge work" is expanding with every release.

I keep coming back to that image from the Dallas Fed research: the door that is not being slammed on anyone, just quietly never opened. It is a gentler form of exclusion, and that is what makes it so dangerous. There are no dramatic headlines about mass layoffs. No protests. No political urgency. Just a generation of qualified, capable young people sending applications into a silence that grows a little deeper each quarter.

The question is not whether AI will transform the labour market. It already has. The question is whether we are willing to notice that the transformation is falling hardest on the people with the least power to do anything about it.