Is AI Really Killing Entry-Level Jobs? New Research Points to Federal Reserve Policy as the Real Culprit

The story has become almost folklore in professional circles: artificial intelligence is systematically eliminating entry-level jobs, creating an entire generation of young graduates locked out of career pathways. Recent college graduates share anxious stories of automated rejection emails and positions that seem to have vanished overnight. The narrative is compelling, intuitive, and according to groundbreaking new research from Google’s Chief Economist team—potentially wrong.

A provocative new paper titled “Looking for the Ladder: Is AI Impacting Entry-Level Jobs?” directly challenges the dominant narrative about AI job displacement. The research, led by Zanna Iscenko (AI & Economy Lead) and Fabien Curto Millet (Chief Economist) at Google, argues that what we’re seeing isn’t the dawn of AI-driven unemployment, but rather the predictable consequences of the Federal Reserve’s most aggressive monetary policy tightening in four decades.

The stakes of getting this diagnosis right are enormous. If we’re attributing cyclical labor market challenges to technological displacement, we risk implementing narrow policy solutions for what may actually be a macroeconomic problem requiring broader interventions.

The AI Job Displacement Narrative

The prevailing narrative emerged from influential research by Brynjolfsson, Chandar, and Chen (2025), dubbed the “Canaries in the Coal Mine” paper. Their analysis found a striking 16% relative decline in employment for early-career workers (ages 22-25) in the most AI-exposed occupations since November 2022. The metaphor was potent: young workers in AI-susceptible roles were the first casualties of an automation wave that would eventually sweep through the broader economy.

This narrative resonated powerfully because it aligned with widespread anxiety about generative AI’s capabilities. Survey data consistently shows that recent graduates express deep concerns about AI’s impact on their career prospects. Media coverage has amplified these concerns, creating a feedback loop where perception and reality become increasingly difficult to separate.

The story seems intuitively correct: AI tools have demonstrated remarkable capabilities in tasks traditionally performed by entry-level knowledge workers. From content creation to basic analysis, from customer service to simple coding tasks, AI appears poised to automate significant portions of junior-level work.

But as compelling as this narrative is, the new research suggests we may be misreading the signals entirely.

The Timeline Problem That Doesn’t Add Up

The first crack in the AI displacement theory appears when we examine the timeline carefully. Job postings for AI-exposed occupations peaked in March and April 2022—a full six to eight months before ChatGPT’s public launch in November 2022. This timing presents a significant challenge to the AI causation theory.

Even more striking, half of the total employment decline observed in the “Canaries” study had already materialized by June 2023. This suggests that within merely six months of ChatGPT’s launch, companies across the American economy not only decided that AI could replace junior staff, but also managed to implement the necessary technological infrastructure, redesigned complex workflows, ensured robust data security, and executed these staffing changes at a national scale.

As the research authors note: “Such a rapid and widespread operational transformation seems implausible.” The timeline becomes even more problematic when we consider that enterprise-grade AI tools weren’t widely available until much later. The OpenAI API launched in March 2023, and ChatGPT Enterprise didn’t arrive until August 2023—well after much of the observed employment decline had occurred.

This timing mismatch suggests that whatever was driving the decline in entry-level hiring in AI-exposed occupations, it wasn’t the sudden availability of consumer chatbots or enterprise AI tools. The cause had to be something that began affecting hiring decisions in early 2022—before most of the business world had even heard of generative AI.

How Fast Are Companies Really Adopting AI?

The timeline problem becomes even more pronounced when we examine actual corporate AI adoption rates, which have been far slower than media headlines might suggest. According to Census data cited in the research, less than 10% of large US businesses planned to use AI for production by Q4 2023. By Q3 2025—nearly three years after ChatGPT’s launch—only 12% of large businesses had actually adopted AI in their operations.

These adoption figures are particularly relevant because they come from large enterprises—precisely the organizations that would be most likely to have the resources and infrastructure necessary for rapid AI deployment. If major corporations with dedicated IT departments and substantial budgets were still struggling to implement AI at scale by 2025, it seems unlikely that the broader economy experienced widespread AI-driven job displacement in early 2023.

The slow adoption rates aren’t surprising when we consider the practical challenges of enterprise AI implementation. Early generative AI models had serious reliability issues, including persistent problems with hallucination and inconsistent output quality. For businesses considering automating any portion of their workforce, these reliability concerns would have been significant barriers to rapid deployment.

Moreover, enterprise AI adoption involves far more than simply subscribing to a consumer chatbot service. Companies need to address data privacy concerns, integrate AI tools with existing systems, train employees on new workflows, and often restructure entire departments. The complexity of enterprise AI implementation makes rapid, economy-wide deployment highly improbable within the timeframes observed in the employment data.

The Real Culprit: Federal Reserve Monetary Policy

If AI wasn’t driving the decline in entry-level hiring, what was? The research points to a much more conventional culprit: the Federal Reserve’s monetary policy. Beginning in March 2022, the Fed initiated its most aggressive interest rate tightening cycle in four decades, raising rates from near zero to over 5% in rapid succession.

The timing alignment is striking. Job postings for high AI-exposure occupations peaked in March-April 2022—precisely when the Fed began its rate hiking campaign. This wasn’t a coincidence; it was cause and effect. As interest rates rose sharply, businesses across rate-sensitive sectors began pulling back on hiring, particularly for roles that weren’t immediately essential to operations.

The mechanism is well-understood in economic literature. Higher interest rates increase the cost of capital, making businesses more cautious about expansion and hiring. Companies postpone non-essential investments, including investments in junior talent that might not generate immediate returns. This effect is particularly pronounced in sectors that are sensitive to interest rate changes—sectors that, as it turns out, employ disproportionate numbers of workers in “AI-exposed” occupations.

The research reveals that 38% of workers in the most AI-exposed quintile are employed in Information, Finance & Insurance, and Professional & Technical Services—all highly interest-rate-sensitive sectors. In contrast, less than 2% of workers in the least AI-exposed occupations work in these same sectors. This sectoral concentration creates a confounding variable problem: when we observe employment declines in “AI-exposed” occupations, we may actually be observing the effects of interest rate sensitivity, not AI displacement.

Why AI Exposure and Rate Sensitivity Are Connected

The connection between “AI exposure” and “interest rate sensitivity” isn’t accidental—it reflects the fundamental nature of knowledge work in the modern economy. The occupations most susceptible to AI automation tend to be white-collar, information-processing roles concentrated in specific sectors that happen to be highly sensitive to interest rate changes.

Consider the types of jobs typically classified as having high AI exposure: financial analysts, content creators, junior consultants, research assistants, and entry-level marketing professionals. These roles are predominantly found in the Information sector (think tech companies and media organizations), Finance & Insurance (banks, investment firms, insurance companies), and Professional & Technical Services (consulting firms, advertising agencies, research organizations).

These same sectors are also among the most sensitive to interest rate changes in the economy. Tech companies rely heavily on venture capital and growth financing, both of which become more expensive as rates rise. Financial services firms see their business models directly affected by rate changes. Professional services organizations often serve clients in rate-sensitive industries, creating downstream effects.

This overlap creates what economists call a confounding variable problem. When we observe employment declines in occupations with high AI exposure, we need to ask: are we seeing the effects of AI automation, or are we seeing the effects of monetary policy transmitted through interest-rate-sensitive sectors that happen to employ workers in AI-exposed roles?

The research strongly suggests the latter. Supporting evidence comes from earlier studies, including work by Zens, Böck, and Zörnerens (2020), which documented the concentration of AI-exposed occupations in interest-rate-sensitive sectors years before the current debate about generative AI emerged.

The Statistical Illusion of Narrow Age Bands

One of the most compelling aspects of the new research is its explanation of why studying narrow age bands (like 22-25 year-olds) can create misleading signals about employment trends. The researchers demonstrate what they call the “mechanical aging” effect—a statistical phenomenon that can make normal hiring slowdowns appear like targeted job elimination.

Here’s how it works: imagine a company that implements a complete hiring freeze but makes no layoffs. In this scenario, the number of workers aged 26 and above would remain relatively stable, as they simply age within their existing roles. However, the number of 22-25 year-old workers would decline dramatically—by approximately 25% per year—simply because they age out of the cohort without being replaced by new hires.

This mechanical aging effect means that any significant slowdown in hiring will disproportionately affect the youngest age cohorts, even if there’s no intentional targeting of young workers or their specific roles. The effect is particularly pronounced for the 22-25 age group because these workers are at the very beginning of their careers, making them most dependent on new hiring to enter and remain in professional roles.

When we combine this statistical artifact with the fact that young workers are typically the first to be affected by economic downturns (a well-documented phenomenon in labor economics), the observed patterns in the “Canaries” data become much less mysterious. The decline in employment for 22-25 year-olds in AI-exposed occupations may simply reflect normal cyclical employment patterns, amplified by the mechanical aging effect in narrow demographic bands.

Supporting this interpretation, the research shows that job postings for both junior and senior positions within high AI-exposure occupations fell roughly in parallel since Spring 2022. If AI were selectively eliminating entry-level work, we would expect to see junior positions declining much more steeply than senior roles. Instead, the parallel decline suggests a broad-based hiring slowdown affecting all levels of employment in these sectors.

Young Workers Always Bear the Economic Brunt

The disproportionate impact on young workers in AI-exposed occupations becomes even less surprising when viewed through the lens of historical labor market patterns. Research by Haltiwanger, Hyatt, and McEntarfer (2018) documents that the “job ladder” is highly procyclical, meaning that career advancement opportunities expand during economic booms and contract during downturns.

Young workers are particularly vulnerable to these cyclical effects because they rely heavily on new job creation and hiring to enter the workforce and advance in their careers. When economic conditions tighten, companies typically freeze hiring long before they begin layoffs. This hiring freeze disproportionately affects recent graduates and early-career professionals who depend on entry-level positions to gain their initial foothold in the professional economy.

Current unemployment statistics bear out this pattern. As of the research publication, unemployment among recent college graduates (ages 22-27) stood at 4.8%, while unemployment for all young workers in the same age group was 7.4%. Both figures are elevated compared to the overall unemployment rate of 4%, demonstrating the familiar pattern of young workers bearing a disproportionate burden during economic stress.

LinkedIn data cited in the research shows that entry-level hiring rates have declined 23% compared to pre-pandemic levels, while overall hiring rates have declined 18%. This 5 percentage point difference illustrates how economic downturns consistently hit entry-level positions harder than the broader job market.

Importantly, the research demonstrates that this pattern of disproportionate impact on AI-exposed occupations isn’t unique to the current period. During the COVID-19 pandemic in 2020—when generative AI couldn’t possibly have been a factor—these same occupations showed steeper employment declines than the broader economy. This historical parallel suggests that AI-exposed occupations have inherent cyclical characteristics that make them more sensitive to economic shocks, regardless of technological factors.

What Other Research Shows About AI Employment

The findings in the Google research aren’t isolated. A growing body of academic literature suggests that fears about immediate, large-scale AI job displacement may be premature. Multiple independent studies are reaching similar conclusions about AI’s current employment effects.

Gimbel et al. (2025) analyzed aggregate US employment trends and found no discernible break in employment patterns following ChatGPT’s launch. If generative AI were causing widespread job displacement, we would expect to see some signal in aggregate employment statistics. The absence of such a signal suggests that AI’s impact on employment remains limited in scope.

Perhaps even more striking is research from Humlum and Vestergaard (2025), who studied AI’s impact on earnings and hours worked in Denmark. Their analysis found what they described as “precise zeros”—no measurable impact on either wages or working hours from AI exposure. Denmark provides an excellent natural experiment for studying AI impacts because of its comprehensive labor market data and relatively rapid technology adoption rates.

These findings align with difference-in-differences analyses that control for other economic factors. When researchers compare AI-exposed occupations to similar but less AI-exposed roles, while controlling for sectoral effects and macroeconomic conditions, the evidence for AI-driven employment declines largely disappears.

This doesn’t mean AI will never impact employment. As the researchers are careful to note, “absence of evidence is not evidence of absence.” The technology is still in its early stages, and its long-term effects remain uncertain. However, the current evidence suggests that immediate, large-scale displacement is not occurring at the pace or scale suggested by the initial “Canaries” research.

Why Getting the Diagnosis Right Matters

The distinction between cyclical employment challenges and technological displacement isn’t merely academic—it has profound implications for policy responses and individual decision-making. If we misdiagnose the current situation, we risk implementing solutions that address the wrong problem while ignoring the actual causes of labor market stress.

Consider the policy implications. If AI displacement is the primary driver of entry-level employment challenges, the appropriate response might include targeted retraining programs, educational curriculum reforms focused on AI-resistant skills, or even policies to slow AI adoption in certain sectors. These interventions would be costly and might not address the underlying economic conditions affecting young workers.

Conversely, if the primary driver is monetary policy and cyclical economic factors, the appropriate response involves different tools entirely: monetary policy adjustments, fiscal stimulus targeted at hiring, or sector-specific support for industries disproportionately affected by interest rate changes. Understanding the root causes of employment trends is essential for crafting effective policy responses.

The research emphasizes that education and training reform is needed regardless of AI’s current impact. The authors note that “the degree to which technology is deployed to automate or augment human labor in an occupation is ultimately a matter of choice—with many different futures being possible at this stage.” This perspective, drawing on work by Autor and Manyika (2025), suggests that how AI affects employment will depend significantly on policy choices and organizational decisions made over the coming years.

For individuals, getting the diagnosis right affects career planning and educational investment decisions. Young professionals who are avoiding certain career paths due to AI displacement fears may be making suboptimal choices based on incomplete information. Similarly, students may be avoiding educational programs in fields that remain viable and important, potentially creating skills shortages in critical areas.

What We Should Monitor Going Forward

While the current evidence suggests that AI isn’t driving immediate job displacement, the researchers outline a comprehensive framework for ongoing monitoring. They emphasize that we’re still in the “early innings” of AI transformation, and vigilance remains essential as the technology continues to evolve.

The proposed monitoring framework includes five key components:

First, monitor both quantities and prices. Employment displacement should affect both the number of jobs available and the wages paid for remaining positions. Current data shows employment changes but no significant wage effects, which is inconsistent with technology-driven displacement but consistent with cyclical hiring slowdowns.

Second, track job postings alongside employment data. Job postings provide early signals about employer intentions and can help distinguish between displacement (where specific roles disappear) and cyclical effects (where hiring slows across all levels).

Third, use multiple measures of AI exposure and usage. Current research relies heavily on occupation-based measures of AI exposure, but actual AI usage varies significantly within occupations and across organizations. Better measurement of actual AI deployment is essential for understanding real impacts.

Fourth, develop conceptual mechanisms linking AI exposure to outcomes. Rather than simply measuring correlations, research needs to identify and test specific causal pathways through which AI might affect employment. This includes understanding how AI changes task composition within jobs, not just whether jobs disappear entirely.

Fifth, watch for “new work” creation. Historical technological transitions have typically created new types of work even as they eliminated others. Monitoring systems need to be sensitive to emerging job categories and changing skill requirements within existing roles.

This monitoring framework reflects the researchers’ balanced approach: acknowledging that current evidence doesn’t support large-scale AI displacement while recognizing that future impacts remain uncertain and worthy of careful observation.

The Bottom Line for Workers and Policymakers

The implications of this research extend far beyond academic debates about causation. For recent graduates and early-career professionals, the findings offer both reassurance and guidance. The current challenges in the entry-level job market appear to reflect familiar economic cycles rather than the beginning of widespread technological unemployment.

This doesn’t mean AI concerns are unfounded. As researcher David Deming notes in related work, technological change remains a significant factor in long-term labor market evolution. However, the evidence suggests that the career ladder isn’t broken by AI—it’s temporarily frozen by economic conditions that will eventually thaw as monetary policy and economic cycles evolve.

For policymakers, the research underscores the importance of evidence-based analysis over reactive policymaking. The temptation to implement dramatic interventions in response to compelling narratives about technological displacement must be balanced against careful examination of underlying data and alternative explanations.

The research also highlights the value of maintaining multiple perspectives on complex economic phenomena. The original “Canaries” research provided valuable insights about employment patterns among young workers, but its attribution of causation to AI appears to have been premature. This evolution in understanding demonstrates the importance of ongoing research and the willingness to revise conclusions as new evidence emerges.

Perhaps most importantly, the research emphasizes that technological change remains a choice rather than an inevitability. As the authors note, “reassurance in the present should not preclude vigilance in the future.” The absence of current evidence for large-scale AI displacement doesn’t guarantee that such displacement won’t occur as the technology continues to develop and mature.

The key insight may be that we’re dealing with a classic case of correlation without causation, complicated by the human tendency to attribute dramatic effects to dramatic new technologies. The Federal Reserve’s interest rate policy may lack the narrative appeal of AI disruption, but it appears to provide a more accurate explanation for current labor market patterns.

As we continue to navigate the intersection of technological change and economic policy, this research provides a valuable reminder that the most compelling explanations aren’t always the most accurate ones. In the case of entry-level employment challenges, the culprit may be found not in Silicon Valley’s AI labs, but in the Federal Reserve’s boardroom where monetary policy decisions continue to shape employment patterns across the American economy.