The Missing Rung

AI Is Removing the Bottom Step of the Ladder.

In April 2026, Goldman Sachs economist Elsie Peng published a research note built not on models or projections but on actual payroll records. The finding was specific: AI has erased roughly 16,000 net US jobs every month over the past year. That compounds to 192,000 jobs in twelve months, removed from real payroll data, measurable now, not forecast for 2030.

The headline provoked the usual debate about whether AI was really causing job losses or whether other factors — tariff uncertainty, Federal Reserve policy, post-pandemic structural shifts — were the real explanation. The debate is legitimate. But it has largely obscured the more important finding buried in the same data, which is not about the total number of jobs lost but about which jobs, held by which workers, are disappearing.

In occupations most exposed to AI substitution, the unemployment rate gap between entry-level workers under 30 and experienced workers aged 31 to 50 has widened sharply relative to pre-pandemic averages. The wage gap has similarly deteriorated, with Goldman’s regression analysis estimating that a one standard-deviation increase in AI substitution exposure widens the entry-level-to-experienced wage gap by roughly 3.3 percentage points.

A week after the Goldman note, Stanford University released its 2026 AI Index. Among its findings: entry-level jobs in software development and customer support have been reduced while mid-career and senior positions have held steady or increased. Among software developers specifically, employment among workers aged 22 to 25 fell nearly 20% since 2024.

Two independent data streams. The same conclusion. AI’s current effect on employment is not evenly distributed across the workforce. It is concentrated at the bottom. The entry-level layer — the apprenticeship rung, the place where workers learn to do the job, build judgment, and become senior workers — is thinning. And the consequences of that thinning extend well beyond the workers it directly affects.

What Entry-Level Work Actually Does

To understand why the missing rung matters, it helps to be precise about what entry-level work is for. It is not primarily about output, at least not in the short term. An entry-level employee costs the organisation more than they produce for the first six to eighteen months of their tenure. The investment is justified because the entry-level role is the mechanism through which organisations reproduce their own capabilities.

Junior workers learn by doing. They handle the volume work — the data processing, the first drafts, the customer queries, the routine analysis — under conditions where their errors are catchable and correctable. They develop professional judgment, which is different from professional knowledge, through the accumulated experience of being wrong in low-stakes situations and correcting their mistakes with the guidance of more experienced colleagues. By the time they become the senior employees who insulate organisations from disruption, they carry institutional knowledge that was not written down anywhere, cannot be onboarded, and cannot be prompted.

This is not romantic. It is structural. The organisations that are now using AI to absorb entry-level work are not simply becoming more efficient. They are making a trade: short-term cost reduction in exchange for a hollowed management pipeline five to ten years from now.

The cruelest part of this shift is structural. Entry-level jobs are likely the ones AI will automate first — and they are what teach young workers to think, build judgment, and ultimately move up. If a company’s bottom rungs are empty, it is hollowing out its own management pipeline years down the road.

The organisations making this trade are, in many cases, doing so without fully accounting for the downstream cost. The quarterly earnings that benefit from reduced junior headcount do not capture the future cost of a senior leadership pipeline populated by people who never learned their craft from the ground up.

The Numbers Behind the Freeze

At companies that adopted AI, junior hiring fell nearly 8% within six quarters — not via layoffs, but through a quiet freeze on new positions, per a Harvard working paper tracking 62 million workers. MarketingProfs

This distinction between layoffs and hiring freezes matters enormously for how the problem is perceived and measured. Layoffs are visible. They generate press releases, severance packages, and unemployment data. A hiring freeze is invisible to the labour statistics. The person who does not get hired does not appear in the unemployment rate as a newly displaced worker. They appear, if at all, as a graduate who took longer than expected to find work, or who accepted a role below their qualification level, or who left the formal labour market entirely.

In the UK, 1.2 million graduates competed for fewer than 17,000 entry-level positions in 2024. Globally, youth unemployment stands at 17% in India, 16.5% in China, 27% in Spain, and 36% in Morocco. Crescendo AI

The US unemployment rate is 4.2%, near generational lows. For recent college graduates aged 22 to 27, it is 5.6%. That remains close to the widest gap on record. Until COVID, college graduates almost always had lower unemployment rates than the overall workforce. MIT Sloan

The inversion of the graduate premium — the historical fact that a degree reliably reduced your unemployment risk — is one of the most significant and least discussed labour market developments of the current period. The labour statistics register it, but the policy conversation has not yet caught up with what it means.

It is also worth being precise about what is driving the freeze, because the AI explanation, while real, is not the complete picture. Federal Reserve Chair Jerome Powell described the current phase as a low-hiring, low-firing economy, driven more by corporate caution in a cooling economy than automation. A Yale Budget Lab study found no discernible disruption in the labour market since ChatGPT’s release in November 2022.

The honest answer is that it is probably both. AI is genuinely suppressing hiring in specific occupations — the payroll data is real and the occupational breakdown is specific. But it is doing so inside a broader economic environment of caution that amplifies its effects. And it is doing so in a way that gives permission to pause hiring — or even shed jobs — in anticipation of capabilities that may not yet have fully materialised. The expectation of AI is shaping hiring decisions as much as the reality.

Who Specifically Is Being Hit

The labour market literature on AI displacement has converged on a finding that is counterintuitive to the narrative of AI threatening manual and physical work. The workers currently most affected by AI’s labour market impact are not factory workers or delivery drivers. They are white-collar, educated, early-career workers in administrative, clerical, and information-processing roles.

Gen Z workers are disproportionately concentrated in the exact types of routine, white-collar, and administrative roles — data entry, customer service, legal support, billing — that AI is best at automating. Without the accumulated experience and specialized judgment that insulate senior workers, they have little buffer against displacement.

Almost 60% of hiring managers use AI as an excuse for layoffs or hiring freezes because it plays better with stakeholders. It suggests that some of the displacement attributed to AI is actually cyclical or structural economic adjustment using AI as its public justification. The practical effect on workers is the same either way. But it complicates the analysis and makes policy responses harder to design, because the problem is not cleanly separable into AI-caused and non-AI-caused components.

The geographic dimension is also significant. Hiring intensity has shifted away from traditional high-cost tech hubs. While cities like Dallas-Fort Worth and Denver have seen hiring declines, secondary markets such as Nashville, Detroit, and Atlanta have shown resilience. This geographic dispersal of opportunity means that the people most affected by the entry-level freeze are not evenly distributed across the country. They are concentrated in the cities and communities that oriented their economic development around attracting the white-collar employers who are now the least likely to be hiring at the entry level.

The Reskilling Gap

The standard policy response to AI-driven displacement is reskilling. Workers whose roles are automated should be trained for the roles that AI creates or augments. The logic is sound in the abstract. The execution has been largely inadequate.

The World Economic Forum estimates that 120 million workers are at medium-term risk of redundancy because they are unlikely to receive the reskilling they need. Despite 82% of enterprise leaders saying their organisation provides some form of AI training, 59% still report an AI skills gap. Only 26% of workers report receiving training on how to collaborate with AI.

42% of employees say their employer expects them to learn AI on their own.

This last finding captures a structural failure in how reskilling is being approached. The assumption embedded in most corporate AI adoption is that the interface is intuitive enough that workers will figure it out. This assumption is wrong in a specific and consequential way. AI tools are intuitive for simple tasks. For the more complex tasks — knowing when to trust AI output, when to override it, how to prompt for useful rather than plausible results, how to integrate AI outputs into workflows that require professional judgment — workers need structured guidance that most are not receiving.

The reskilling gap is not primarily a motivation problem. In the US, 70% of workers surveyed said they completed AI training when their employers made it available. The problem is availability, not willingness. And availability is an institutional choice, not a market outcome.

The Seniority Cliff

There is a second-order consequence of the missing rung that the current labour market data cannot yet capture but that researchers are beginning to model: the seniority cliff.

If entry-level hiring freezes persist for three to five years, the organisations that have frozen hiring will face a simultaneous shortage of experienced workers at the point when those workers would normally have been promoted from the entry-level cohort. The people who would have been junior employees in 2024 and 2025 — learning their craft, building institutional knowledge, developing the professional judgment that makes senior workers valuable — did not get hired. They are not in the pipeline. When the AI capabilities that justified the freeze plateau or prove less reliable than projected, organisations will need experienced workers they did not train.

The most significant second-order insight derived from this analysis is the risk of a seniority cliff in the next five to ten years. Policy discussions now focus on retooled jobs — occupations where the title remains the same but the skills change entirely due to AI. This requires a shift in funding from traditional higher education to continuous, skills-based certification.

The seniority cliff is not a prediction that should be treated as certain. AI capabilities may continue to improve in ways that eliminate the need for junior workers permanently, not just temporarily. The new roles that AI creates may provide alternative pathways to professional development that do not require the traditional apprenticeship model. The labour market is adaptive, and workers are adaptive within it.

But the adaptation is not automatic, and it is not free. It requires deliberate investment in the alternative pathways — the apprenticeship programmes, the skills-based certifications, the structured entry points that do not require passing through a traditional entry-level employment role. Most organisations have not made that investment. Most governments have not required it. And most workers are navigating the transition without the institutional support that previous generations received when the bottom rung was still attached to the ladder.

What This Means for Education

The entry-level freeze has specific implications for education systems that have organised themselves around the proposition that a degree is a reliable bridge between formal education and stable employment.

The Stanford 2026 AI Index finding on STEM graduate unemployment is worth dwelling on. Computer engineering graduates facing 7.5% unemployment and computer science graduates at 6.1% — both above the national average — represents the collapse of a narrative that has been used to justify tuition costs, immigration policy, and economic development strategy for two decades. The STEM premium did not disappear. It shifted. The premium now attaches not to the credential itself but to the specific skills and experiences that allow workers to work with AI rather than be replaced by it.

The education system has not caught up with this shift. Universities are still primarily preparing students for the entry-level roles that the labour market is contracting. The skills that would protect graduates — understanding how AI systems work, when to trust and when to override AI output, how to do the creative and relational and judgment-intensive work that AI cannot yet do — are not systematically integrated into most degree programmes.

Gartner’s strategic predictions warn that atrophy of critical-thinking skills due to GenAI use will push 50% of organisations to require AI-free skills assessments by 2026. This is the education system’s paradox made visible. Students are learning to use AI as a tool for producing outputs that look like the work they are being educated to do. Employers are increasingly aware that the outputs and the underlying competency are not the same thing. The credential that was supposed to signal competency is being undermined from both ends.

The Policy Gap

The policy response to the missing rung has been inadequate in predictable ways. Labour statistics that measure unemployment rather than hiring rates are slow to register a freeze. Reskilling programmes that require institutional delivery are not scaling at the speed of AI deployment. Education policy that moves on decade-long cycles cannot respond to a labour market that is transforming on a two-year cycle.

The most promising responses are also the most unglamorous. Apprenticeship programmes that create structured entry points not dependent on traditional employment. Skills-based certifications that decouple credential from institutional affiliation. Procurement requirements that tie government AI contracts to commitments on junior hiring. Disclosure requirements that make AI’s contribution to hiring decisions legible in labour market statistics.

None of these are sufficient on their own. But they share a common logic: the missing rung will not be replaced by the market spontaneously, because the market incentives that removed it — cost reduction, efficiency, the ability to substitute AI for entry-level labour — remain in place. Replacing the rung requires deliberate institutional design. And the window for designing it before the seniority cliff arrives is shorter than most of the people who would need to design it currently appreciate.

References

Peng, E. (2026, April 6). AI substitution and augmentation in the U.S. labor market. Goldman Sachs U.S. Daily Note.

Stanford HAI. (2026, April 13). AI Index Report 2026. Stanford University Human-Centered Artificial Intelligence.

Anthropic. (2026). Labor market impacts of AI: A new measure and early evidence. https://www.anthropic.com/research/labor-market-impacts

Fortune. (2026, April 6). AI is cutting 16,000 U.S. jobs a month — and Gen Z is taking the brunt, Goldman Sachs says. https://fortune.com/2026/04/06/ai-tech-displacement-effect-gen-z-16000-jobs-per-month

Axios. (2026, April 21). Young people hate today’s job market. You can’t blame it all on AI. https://www.axios.com/2026/04/21/gen-z-jobs-unemployment-college-grads-ai

Axios. (2026, April 16). Gen Z is terrified of the AI revolution. Nobody’s preparing them for it. https://www.axios.com/2026/04/16/ai-use-gen-z-college-jobs-fear

World Economic Forum. (2025). The rising pressures for Gen Z in the global job market. https://www.weforum.org/stories/2025/11/gen-z-labour-market-ai-economy

World Economic Forum. (2026, January). AI perception gap: How to ensure employers and workers are ready for transformation. https://www.weforum.org/stories/2026/01/ai-perception-gap

Rezi.ai. (2026, January). The crisis of entry-level labor in the age of AI 2024-2026. https://www.rezi.ai/posts/entry-level-jobs-and-ai-2026-report

Iternal.ai. (2026). AI skills gap: Statistics, causes and how to close it. https://iternal.ai/ai-skills-gap

Gloat. (2026). 10 key AI workforce trends in 2026. https://gloat.com/blog/ai-workforce-trends