A labor market can weaken without beginning with mass layoffs. It can start when a worker who has almost always been employed discovers that the next job is no longer easy to find. That loss of an exit route matters before it appears as a large unemployment stock: it changes whether employees switch firms, negotiate pay, tolerate restructuring or remain in a role that no longer fits.
A new Richmond Federal Reserve analysis identifies precisely that pattern. The finding is especially relevant to white-collar and technology-linked businesses, but it needs a skeptical reading. The data show an unusual divergence and a plausible AI channel. They do not yet prove that AI caused the decline.
The warning sits in a flow, not the unemployment rate
The unemployment rate is a stock: the number of people classified as unemployed relative to the labor force at a point in time. A job-finding rate is a flow: the probability that an unemployed person moves into employment. Stocks can look stable while the transitions underneath them become less favorable.
The Bureau of Labor Statistics explains that these flows can be estimated because roughly three-quarters of households in the Current Population Survey are interviewed in consecutive months. Its labor-force flow series tracks movement among employment, unemployment and nonparticipation. That approach reveals churn hidden by a small net change. It is also survey evidence, with classifications and statistical adjustment, rather than a direct census of vacancies or hiring decisions.
The Richmond Fed researchers add a model-based classification. Primary workers, about 55% of the population in their framework, are almost always employed. Secondary workers, about 14%, are strongly attached to work but experience unemployment more often; despite their smaller population share, they account for most unemployment and its normal cyclical movement. A deterioration among primary workers can therefore be economically important even while they remain a small part of the unemployment pool.
Stable workers lost more of their outside option
From the November 2022 peak to the September 2025 trough, the estimated job-finding rate for primary workers fell 13 percentage points. The rate for secondary workers fell only 2 points. The researchers say that gap is unlike previous recessions, when all groups weakened cyclically and the secondary segment continued to carry most of the adjustment.
The immediate inference is not that stable workers suddenly became the most unemployed. They did not. It is that their historical advantage in returning to work narrowed sharply. Business Insider's account of the research describes the practical consequence: experienced workers seeking another role face a more daunting search.
That changes an employee's outside option — the credible alternative to staying with a current employer. When the option weakens, voluntary mobility can fall and wage bargaining can soften without a wave of dismissals. For companies, that may ease near-term compensation pressure and reduce unwanted turnover. It can also create a slower, less efficient matching process in which growing firms struggle to attract precisely the workers who are reluctant to risk a move. Lower churn is not automatically higher productivity.
AI exposure is a correlation with a mechanism
The technology result is sharper but less causal than a headline may imply. The researchers rank occupations by overlap between their tasks and capabilities described in AI patents. Computer programmers, financial analysts and engineers sit among the more exposed roles; construction, food service and personal-care work are less exposed. Before 2023, job-finding rates across exposure groups moved together. Since then, highly exposed occupations have experienced the largest declines.
There is a plausible mechanism. If AI allows an existing team to produce more, employers may fill fewer incremental positions or demand a broader skill set from each hire. A well-known field study, Generative AI at Work, found that an AI assistant raised customer-support productivity by 14% on average in its setting, with a 34% improvement for novice and lower-skilled agents and little effect on the most experienced workers. That is evidence that technology can change the productivity distribution inside a job. It is not evidence that the same percentages apply across the economy or that headcount must fall.
Alternative explanations remain strong. AI-exposed jobs are concentrated in sectors that experienced pandemic-era hiring, later cost discipline and sensitivity to financing conditions. Employers may be cautious because demand is uncertain, budgets are being reset or prior hiring ran ahead of revenue. The exposure index may partly identify those sectors rather than isolate a technology shock. The Richmond Fed authors explicitly leave aggregate versus technology-specific attribution to further work.
Hiring and wage behavior can separate the stories
The distinction matters for investors. A broad low-hire environment should depress vacancy creation and job-finding across many occupations as demand softens. A technology-specific channel should remain concentrated in highly exposed roles after controlling for industry, employer growth, interest sensitivity and prior overhiring. It should also appear in the composition of openings: fewer entry roles, wider task requirements or more output per worker where tools are deployed.
Corporate evidence can test the mechanism. Investors should compare disclosed headcount with revenue and workload, track whether AI adopters reduce external hiring or simply redeploy staff, and examine wage growth and job-to-job transitions in exposed occupations. Rising productivity alongside stable output and persistently weaker hiring would strengthen the substitution case. A rebound in exposed hiring when budgets recover would favor the cyclical explanation.
The evidence that would change this analysis is therefore not another executive prediction about jobs. It is a matched pattern across adoption, vacancies, hires, wages and output, ideally at the firm and occupation level. Until then, the Richmond Fed finding is best treated as a warning about mobility. America's most stable workers appear less able to convert experience into a quick new match. AI may be part of the reason, but the data have not yet separated the machine from the hiring cycle.