Updated: October 1, 2025
Theoretical predictions unmet. If AI truly functions as a knowledge system, it should increase the ratio of junior to senior labor by lowering the cost of knowledge retrieval and decision-making. Yet statistics suggest the opposite. Whether this reflects a temporary lag or a deeper theoretical shortcoming remains an open question.
Empirical evidence inconclusive. Current studies point in both directions: some find declining demand for junior labor and rising demand for senior roles, while others suggest the reverse. The data thus far do not resolve the debate about AI’s true impact on workforce composition.
Future of labor markets. The balance between junior and senior roles will depend on how AI systems are designed and deployed — whether as knowledge aggregators, knowledge generators, task executors, or hybrids. These shifts will unfold alongside broader transformations in organizational structures.
Figure 1. Junior Headcount (Blue Line) Decrease among Software Developers according to Brynjolfsson et al. (2025)
Over the past several months, I have immersed myself in studies examining how artificial intelligence (AI) is reshaping the labor market. Some findings point to an increase in junior roles within firms adopting AI, while the dominant storyline warns of a sharp decline in entry-level opportunities. This doomsday framing often paints a picture of a system where the pipeline of fresh talent is slowly drying up and the demand for seniors is nowhere to be fulfilled.
One particular study from the 2000s caught my attention, which explained the role of information systems in the composition of the labor force. It prompted me to reconsider the logic behind how reductions in the cost of knowledge retrieval due to AI might alter labor dynamics. This blog reflects on that question by revisiting the theoretical predictions about how information and communication technologies, AI included, shape the demand for junior workers.
Figure 1. Dual Nature of AI Knowledge Systems
Role of AI as an information system can be understood from two perspectives (see Figure 1). First, they can be viewed as information systems that aggregate and synthesize knowledge from diverse sources (as I already wrote about here). Second, they can be seen as intelligent agents capable of generating qualitatively new insights that exist independently of human input. While Agarwal et al. (2025) adopt the latter, future-oriented perspective, this discussion focuses on the current state of affairs, where AI functions primarily as an aggregator of information accessible to a wide range of firms at a reasonable investment cost.
Figure 2. Empirical Evidence of Both Decrease and Increase of Junior Labor
Building on this framing, earlier work provides useful theoretical background and predictions. Garicano (2000) predicts an increase in production workers — namely, employees with the most common and easiest knowledge (aka juniors) — as a result of technological advancements in information systems. By this logic, the introduction of an expert system, read AI, would increase the scope of decision-making by lower-level workers, which consequently increases the ratio of production workers to problem solvers. Hence, if the ratio increases, the number of "junior" production workers should increase. Similarly, Babina et al. (2023) provide empirical evidence of growth in low-rank positions (see Figure 2). These findings suggest that AI adoption should, in principle, increase demand for junior workers, who offer cost advantages compared to more senior staff. However, this prediction appears to contradict other pieces of evidence (see Figure 2) and much of the current media narrative, which frequently highlights the displacement of entry-level roles.
Figure 3. AI Systems' Dual Purpose
This tension between theoretical prediction and empirical evidence raises the possibility that firms may already employ a sufficient number of junior workers, leaving limited scope for further expansion in such roles. Alternatively, other forces may be shaping employment dynamics. One explanation is that generative AI (GenAI) is perceived less as a tool that reduces the cost of knowledge acquisition and more as a straightforward task automation system (see Figure 3).
At this stage, both interpretations remain plausible, and the question of which mechanism more accurately describes current labor market dynamics requires further investigation, both conceptual and empirical.
Practically speaking, the definition of AI as one or the other is not deterministic and hugely relies on the design of these systems. A little time has passed to make the final judgement, and limited evidence is presented to claim the stability of the discovered effects on the labor markets.
I note down the questions below to think about for future research and reflect as new types of AI applications enter our desktops:
Nature of AI Systems. Are they knowledge systems, executive & automation systems, or hybrids? Careful consideration should be given to the context and circumstances, e.g., industry, domain, in which this assessment is being made.
Nature of Jobs. How to sustainably classify the jobs as AI vulnerable or not? Such classification should necessarily rely not only on the task displacement effect by AI, but also on the task reinstatement effect, according to Acemoglu.
Business Considerations. How do companies actually think about AI systems? The answer to this question should closely account for the tradeoffs between cost-saving on labor and investment cost in technology capabilities, especially if they are provided by the monopolist.
Wider Economic Considerations. Is the model of 'AI Geniuses' sustainable in the long, long run? I am considering the impact on the consumption dynamic when the labor income is gone.
The resemblance of AI to earlier generations of information systems is striking. Much like their predecessors, today's AI systems aggregate knowledge and make it broadly accessible. What has changed is the level of refinement: AI reduces search costs more effectively and delivers conditional, context-sensitive outputs tailored to the user. In this sense, AI represents not a fundamental break but a continuation of the long tradition of information systems development, where each iteration has steadily driven down the cost of retrieving knowledge. At the same time, these systems have also gained modest but meaningful capabilities in automation — still imperfect, but broader in scope than previous waves, and therefore consequential.
Whether we classify AI as a knowledge system or as an automation system, the theoretical expectation is that demand for junior labor should rise. Garicano’s framework suggests that empowering decision-making at lower levels reduces costs and should expand opportunities for entry-level workers. Acemoglu’s theory of the reinstatement effect likewise implies that each wave of automation creates demand for new skills, potentially reshaping but not eliminating junior roles. Yet, in practice, the market has failed to follow these predictions. A creative explanation may be that the new skillsets now demanded are tied to a higher baseline of knowledge, leading to two logical implications: Firms treat AI as primarily an automation system, and organizations remain structurally overleveraged with hierarchical layers that limit openings at the bottom.
Ultimately, no development path forward is predetermined. If the theories are correct, the junior job market may rebound as firms adapt. If they are not, there are still avenues to convert “potential juniors” into a productive economic force — through education that equips them with complementary skills and reinvents how the job tasks are done (synthetic reinstatement), and through entrepreneurship that allows them to bypass hierarchical bottlenecks altogether. In either case, the future of junior work will be shaped not only by the technology itself but by how firms and institutions choose to integrate it.
Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3–30.
Babina, T., Fedyk, A., He, A. X., & Hodson, J. (2023). Firm investments in artificial intelligence technologies and changes in workforce composition (NBER Working Paper No. 31325). National Bureau of Economic Research.
Brynjolfsson, E., Chandar, B., & Chen, R. (2025). Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence (Stanford Working Paper). Stanford University.
CNBC. (2025, September 7). AI entry-level jobs: Hiring, careers. CNBC. https://www.cnbc.com/2025/09/07/ai-entry-level-jobs-hiring-careers.html
Garicano, L. (2000). Hierarchies and the Organization of Knowledge in Production. Journal of Political Economy, 108(5), 874–904.
Please cite this article as:
Petryk, M. (2025, October 1). The Hiring Enigma: Why AI Growth Isn’t Creating the Junior Roles We Expected. MariiaPetryk.com. https://www.mariiapetryk.com/blog/post-24