Generative AI is reorganizing the job ladder, with junior and senior roles changing differently

Generative artificial intelligence (GenAI) is not only threatening to displace exposed jobs but also pushing firms to reorganize labor demand by shifting hiring away from some roles while rewriting the tasks inside others, reveals a new study published as an arXiv economics working paper.

Based on more than 9.3 million U.S. job postings from 2021 to 2025, the paper titled "Generative AI and the Reorganization of Labor Demand" finds that exposure to generative AI is dynamic, varies across the job ladder and reflects a deeper redesign of work rather than a simple decline in AI-exposed occupations.

AI exposure is changing inside the labor market, not just across occupations

The study challenges a major assumption in much of the current debate over AI and jobs: that exposure to generative AI can be treated as a fixed feature of an occupation. Most earlier studies have tried to identify which occupations are most exposed to large language models and then asked whether those occupations are losing jobs or hiring demand. The new analysis states that this approach misses an important shift. Employers are changing the task content of jobs as the technology spreads.

The research uses a nationwide dataset of online job postings from Lightcast, covering all major sectors of the U.S. economy from January 2021 through June 2025. The raw data include more than 188 million postings, from which the study draws a structured sample of 9,373,092 postings. The sample preserves variation across occupation, industry, seniority and time, allowing the analysis to track how job content changes across the labor market.

To measure generative AI exposure, the study uses a two-stage large language model pipeline. First, it extracts the specific tasks described in each job posting rather than relying on generic occupation-level task lists. Thereafter, it classifies each task by whether current generative AI tools could substantially reduce the time required to perform it while maintaining comparable quality. These task-level classifications are then combined into a posting-level exposure measure.

The design matters because two jobs with the same occupation title may contain very different responsibilities. A marketing specialist in finance, a marketing specialist in retail and a marketing specialist at different seniority levels may face different levels of AI exposure. A fixed occupation-level score cannot capture those differences. A posting-level measure can.

The findings show that generative AI exposure rose through early 2022, declined through 2023 and partially recovered afterward. The pattern indicates that AI exposure does not remain stable as firms adjust hiring and task requirements. Exposure also differs sharply across sectors. Finance and insurance, professional services and information have the highest average exposure, reflecting the concentration of language-intensive, analytical and digital work in those industries. Accommodation and food services, retail trade, transportation and warehousing sit at the lower end because their postings involve more physical, in-person and operational tasks.

The study also finds a clear seniority gradient. Senior postings have the highest exposure on average, followed by junior postings and then intermediate roles. This complicates the common view that entry-level work is always the main target of generative AI disruption. Junior jobs remain important, but senior roles often include decision-making, analysis, communication and information-processing tasks that are highly exposed to AI assistance.

Firms are reallocating hiring and redesigning job tasks

The paper decomposes labor-demand adjustment into two channels. The first is hiring reallocation, where firms change the mix of jobs they post and the other is job redesign, where firms keep posting similar jobs but alter the tasks inside those jobs. If firms only adjust through reallocation, the central question is which occupations or industries are gaining and losing postings. However, if firms also redesign roles, then the effect of AI is deeper, which means the content of work is changing even when the job title remains the same.

The study finds that both channels are active. After the third quarter of 2023, hiring reallocation accounts for 52.01% of the aggregate decline in generative AI exposure. Within-job task redesign accounts for 39.46%, while the interaction between the two accounts for 8.54%. In other words, the decline in exposure is not just a matter of fewer postings in highly exposed occupations. A large share comes from employers changing the task structure of jobs they continue to advertise.

The shift becomes visible after generative AI tools moved from public experimentation toward wider enterprise deployment. The study does not claim that a single product launch caused the turning point. But the timing is consistent with firms beginning to integrate generative AI into organizational workflows and then changing how they describe labor needs.

The job redesign channel is especially important because it is invisible in many traditional labor-market measures. Employment data can show whether workers are hired or displaced. Occupation-level exposure scores can show which jobs appear vulnerable. But neither can show whether firms are rewriting the tasks, skills and responsibilities attached to existing job categories.

The paper also uses a separate Oaxaca-Blinder decomposition to identify which observable job characteristics explain the post-GPT exposure decline. Occupational shifts account for about 90% of the exposure change attributable to observed job characteristics. This confirms that firms are shifting labor demand across occupations. But other factors also matter, including remote-work status, industry, employment type and internship status.

Remote work plays a notable role. Remote postings were more exposed to generative AI before the post-GPT period, and the share of remote postings declined afterward. That shift reduced aggregate exposure. The pattern suggests that some changes in AI-exposed labor demand may be tied not only to AI adoption, but also to broader shifts in workplace arrangements, including return-to-office trends.

The study also addresses concerns that AI exposure may be confused with macroeconomic effects. Highly exposed sectors, such as finance, information and professional services, may also be sensitive to interest rates and wider economic slowdowns. To test this, the study repeats its analysis within industries, removing cross-sector shifts from the calculation. The main pattern remains: firms adjust through both reallocation and task redesign even within sectors.

The job ladder is being reorganized in different ways

The study finds that senior, intermediate and junior jobs all adjust to generative AI, but they do so through different mechanisms. Senior jobs adjust earlier and mainly through hiring reallocation. From the third quarter of 2023 onward, the composition effect accounts for 70.80% of the aggregate contribution among senior postings. This means firms appear to alter the structure of senior vacancies first, shifting away from senior job cells that were more exposed to generative AI at baseline. Task redesign appears later and plays a smaller role for senior roles.

Junior jobs follow a broader pattern. Hiring reallocation remains the largest factor, accounting for 60.15% of the aggregate contribution among junior postings. But task redesign and the interaction between reallocation and redesign also matter. This suggests that entry-level work is being changed along multiple margins at once. Firms are not only altering the kinds of junior jobs they post. They are also changing the tasks embedded in those jobs.

Entry-level jobs often serve as training grounds where workers learn skills, build judgment and move into more advanced roles. If generative AI changes the tasks available in junior positions, it may affect not only current hiring but also the structure of career progression. Workers entering the labor market may face fewer opportunities in some categories and a different mix of responsibilities in the roles that remain.

Intermediate jobs track the overall labor market most closely. Reallocation and redesign contribute almost equally, indicating that mid-level roles are absorbing AI-driven changes through both vacancy composition and task structure.

In a nutshell, the study states that generative AI is reorganizing the architecture of work. The labor-market response is not captured by a simple substitution story in which exposed jobs disappear. Nor is it fully captured by an augmentation story in which the same jobs remain but workers become more productive. The evidence points to a third process: organizational reconfiguration.

The study's limitations also point to the next research frontier. Job postings capture employer demand at the moment of hiring, but they do not directly show what workers do after they are hired or how incumbent employees experience AI-driven redesign. Future research linking job postings to employer-employee data could reveal how these shifts affect wages, employment stability, promotions and career mobility.