A few weeks ago, we participated in a meeting of faculty members and researchers discussing the integration of generative AI (GenAI) into academic work. As expected, the conversation revolved around opportunities, challenges, productivity gains, teaching, research, and the future of the university.
Then, one faculty member made a comment that immediately caught our attention. “Recently,” he said, “I realized that new graduate students slow my work down.” Several other faculty members nodded in agreement and expanded on this idea. “I found that we can work without graduate students,” one remarked. Another added, “Students need to learn how to work with these tools so they can accelerate our research process.”
None of these comments were intended to criticize graduate students. Rather, they reflected a new reality emerging in the age of GenAI. Tasks that were once delegated to graduate students—searching the literature, summarizing papers, drafting text, analyzing data, and even generating research ideas—can now be completed at a much faster pace with the assistance of GenAI tools.
As we listened to the discussion, something suddenly clicked in our minds. For months, we have been reading and researching the ‘junior developer crisis’ in software engineering and its implications for computer science education. We knew that companies were hiring fewer entry-level software engineers because AI could perform many tasks traditionally assigned to junior developers. However, during the discussions following this meeting, we realized that this phenomenon was not confined to computer science and that it could enrich our understanding of the changes needed in computer science education.
We asked, what if the issue is much broader? What if we are witnessing the emergence of a junior professional crisis that extends across academia, law, consulting, finance, and many other knowledge-intensive professions? Specifically, among graduate students in academia, junior engineers in the industry, junior lawyers in law firms, junior consultants in consulting companies, and junior analysts in finance. In profession after profession, GenAI is beginning to perform many of the tasks that traditionally served as training grounds for newcomers.
This observation led us to a deeper question: If AI increasingly performs the work through which novices learn, how will future experts emerge?
This blog post attempts to explore this question.
When Machines Perform the Learning Tasks
Much has been written about the impact of GenAI on software engineering (for example, Alex Wright’s “The Vanishing Apprentice”). Headlines warn that junior programmers are struggling to find jobs because AI can now generate code, write tests, and debug software. With the rise of agentic coding, AI systems can now perform tasks that previously required teams of software developers to collaborate. However, focusing only on computer science may cause us to miss a much larger phenomenon. What appears to be a crisis for junior developers may actually be the first manifestation of a broader transformation affecting many knowledge-intensive professions in the future.
The challenge is not merely technological. It concerns how expertise develops in society.
For generations, most professions have followed a similar pattern of expertise development. Junior professionals begin by performing relatively simple and repetitive tasks. Although these tasks are often viewed as tedious by experienced practitioners, they serve an essential educational purpose. Through observation, repetition, and feedback, novices gradually acquire experience, internalize the knowledge, judgment, and professional norms of their field, and eventually become experts.
Thus, junior software engineers reviewed code, fixed bugs, implemented new features, tested functionality, and supported senior engineers with implementation, testing and debugging. Junior researchers collected data, reviewed the literature, organized references, and conducted preliminary analyses. Junior lawyers reviewed documents, summarized cases, conducted legal research, and prepared draft documents. Junior consultants prepared presentations, analyzed datasets, and gathered information for senior consultants. Junior accountants reconciled records and prepared routine reports. Although these tasks may appear mundane, they provide the practical training through which professional expertise is developed.
Today, GenAI is changing this equation. GenAI is beginning to automate many of the tasks that traditionally served as training grounds for newcomers. Large language models can summarize documents, draft reports, generate code, search the literature, prepare presentations, review contracts, and produce initial analyses within seconds.
As a result, organizations may achieve short-term productivity gains while inadvertently weakening the pipeline through which future experts are developed. From a managerial perspective, the appeal is obvious: why assign a junior employee a task that requires several hours when an AI system can complete most of it in minutes?
The economic incentives are compelling. Organizations facing budget constraints and competitive pressures naturally seek greater efficiency. As AI tools continue to improve, managers may increasingly rely on them to perform work that was once assigned to entry-level employees.
The result is a paradox. The very tasks that organizations seek to automate are often the tasks through which novices traditionally develop into experts.
Beyond Computer Science
As AI can already perform many tasks traditionally assigned to junior developers, software engineering may represent the most visible manifestation of this broader trend. This is partly because the field provides vast amounts of structured digital training data enabling AI systems to achieve high levels of performance. As a result, the junior-role crisis has emerged first in software engineering, although similar dynamics are beginning to appear in other professional domains.
Research has traditionally involved a long apprenticeship process. Graduate students and junior researchers spend years reading articles, synthesizing bodies of knowledge, designing studies, analyzing data, and drafting manuscripts. These activities have not only contributed to research productivity but have also served as essential mechanisms through which expertise is developed.
Academic Research
Today, AI can perform many of the tasks that have traditionally constituted the training ground for novice researchers. It can rapidly summarize large bodies of literature, suggest research questions, suggest statistical methods, draft manuscripts, and organize references.
These capabilities undoubtedly enhance research productivity. However, they raise a fundamental question: If AI performs much of the reading, summarizing, and drafting, how will novice researchers develop the deep understanding that traditionally emerges through sustained engagement and struggling with the literature?
Many experienced researchers can recall formative periods during which they read hundreds of articles, manually compared competing arguments, identified conceptual connections, and gradually developed an intuitive understanding of their field. These experiences were time-consuming, yet they were often intellectually transformative because expertise emerged through this process.
If AI substantially shortens this journey, what are the implications for expertise development? Were these labor-intensive experiences essential for becoming an expert, or were they merely the historical pathway through which expertise was acquired? If the latter, how should we redesign educational and professional experiences to enable students to develop expertise in an AI-mediated research environment?
Law
The legal profession provides another compelling example of this dynamic. For decades, junior associates developed their expertise by reviewing contracts, researching precedents, summarizing cases, and drafting legal documents. These activities were often routine and demanding, yet they served as essential training mechanisms through which professional expertise was gradually developed.
Today, AI systems can perform much of this work at unprecedented speed and scale. Law firms may achieve significant efficiency gains; however, they may also encounter a long-term challenge: Where will the next generation of senior attorneys come from if junior professionals have fewer opportunities to develop their expertise through practice?
Legal expertise involves far more than retrieving relevant cases or generating documents. It requires nuanced interpretation, strategic thinking, ethical reasoning, and understanding of complex human contexts. These capabilities emerge gradually through sustained exposure to real cases and professional experience.
The fundamental question is therefore not whether AI will replace lawyers. Rather, it is whether AI will transform the process through which lawyers become experts.
Consulting and Business Analysis
The same pattern is emerging in consulting and business analytics. Consulting firms have traditionally relied on large numbers of junior analysts who gather information, prepare presentations, conduct market research, and analyze data. These activities, although often routine, have served as the foundation through which consultants develop analytical skills, business judgment, and professional expertise. Many of these tasks can now be performed, or substantially accelerated, by AI systems, reducing the demand for entry-level work and disrupting traditional early-career pathways.
With the help of AI, a consultant can now create a first draft of a market analysis, competitor review, or strategic presentation in a fraction of the time previously required. This creates significant productivity gains but may simultaneously reduce opportunities for junior analysts to acquire foundational skills that support future expertise. Again, the issue is not merely job replacement alone. It is the transformation of the pathways through which professionals develop expertise.
The Broken Bridge to Expertise: AI and the Future of Professional Learning
One possible consequence of the increasing replacement of junior roles by AI systems is what might be called the ‘missing middle’ phenomenon: the disappearance of the developmental stage between novice and expert.
Historically, organizations have maintained a pipeline progressing from junior through intermediate to senior roles. AI may weaken the first stage of this progression. If fewer junior professionals are hired, fewer professionals will have opportunities to gain experience. Years later, organizations may face a shortage of mid-level and senior professionals.
This is ultimately a question about the future of expertise itself. Will AI undermine the traditional mechanisms through which expertise is cultivated and transmitted across generations? Will it enable new pathways through which experts can be developed more efficiently? The answer may shape the future of many professions.
History offers useful parallels, while also highlighting what makes GenAI different.
When calculators became widespread, some feared the decline of mathematical thinking. Instead, education adapted. Students continued to learn arithmetic, but the emphasis gradually shifted toward higher-level reasoning and problem solving.
Similarly, computer-aided design transformed engineering without eliminating engineers. Instead, it automated routine drafting tasks and shifted engineering work toward higher-level design, analysis, simulation, system thinking, and decision-making. In this sense, the profession evolved rather than disappeared, as new tools reshaped both the nature of engineering tasks and the skills required to perform them.
GenAI may trigger a similar transformation. However, there is an important difference: previous technologies primarily automated calculations and repetitive physical or procedural tasks. GenAI, in contrast, automates cognitive activities that have traditionally functioned as learning experiences for knowledge workers.
The challenge, therefore, is to redesign pathways through which individuals develop expertise in an AI-mediated world.
Rethinking Professional Education
If traditional apprenticeship models weaken, educational institutions may need to assume new responsibilities. Universities have long relied on workplaces to provide authentic professional experiences after graduation. In a GenAI-rich world, however, this assumption may no longer hold. Consequently, students may need more opportunities to engage with the kinds of challenging tasks that organizations increasingly automate instead of assigning them to junior employees.
Professional education may therefore need to place greater emphasis on deliberate practice1—structured learning experiences that intentionally develop judgment, reasoning, and decision-making, through complex problem-solving, interdisciplinary thinking, critical evaluation of AI-generated outputs, effective human-AI collaboration, feedback, and timely reflection.
That is, rather than preparing students primarily for entry-level tasks, higher education may need to redesign learning experiences around the development of expertise itself. This shift could reshape educational programs across fields such as computer science, law, medicine, business, engineering, and other disciplines.
The Future of Expertise and Its Formation
Perhaps the deeper transformation concerns the nature of expertise itself. Traditional expertise involved mastering and performing tasks directly. Future expertise may increasingly involve supervising, evaluating, and directing AI systems. The expert of tomorrow may spend less time producing professional documents and code, and more time validating them; less time searching for information and more time interpreting it; less time performing routine analyses and more time deciding which analyses are meaningful and actionable.
This transition does not eliminate the need for expertise. On the contrary, it may increase its importance. When GenAI can generate many plausible answers, human judgment becomes even more critical and valuable. The challenge is ensuring that future professionals develop this judgment despite having fewer opportunities to perform the activities through which previous generations acquired it.
This perspective suggests that public discussions, which often focus on whether AI will replace jobs, should be expanded to address a deeper educational question: How will future expertise be developed? Should organizations preserve traditional apprenticeship activities as part of professional training, or should they redesign learning pathways to cultivate expertise in an AI-mediated world?
This issue extends far beyond software engineering. It affects professionals across a wide range of knowledge-intensive fields, including research, law, consulting, accounting, and journalism.
This is a societal challenge. While organizations are understandably focused on productivity gains, societies must also consider the sustainability of expertise pipelines. The professionals who will make critical decisions ten or twenty years from now are today’s students and junior professionals. If AI transforms or eliminates traditional pathways of expertise development, new educational pathways will need to be intentionally designed.
The central challenge of the AI era may therefore lie in ensuring that the next generation has meaningful opportunities to become experts in the first place. In this sense, the junior crisis is fundamentally a challenge of professional formation, not simply a challenge of employment. As such, it represents one of the most important educational and societal challenges emerging from GenAI. Addressing this challenge requires academia to reconsider how to design learning experiences that enable students to develop professional judgment and expertise in an environment where many traditional entry-level tasks are increasingly automated.
References
1. Ericsson, K. A., Krampe, R. T., and Tesch-Römer, C. The role of deliberate practice in the acquisition of expert performance. Psychological Review 100, 3 (1993), 363–406. https://doi.org/10.1037/0033-295X.100.3.363
Orit Hazzan is a professor at the Technion’s Faculty of Education in Science and Technology. Her research focuses on computer science, software engineering, and data science education. Additional details about Hazzan’s professional work can be found on her website; her email is oritha@technion.ac.il.
Yael Erez is a lecturer at the Technion’s Faculty of Computer Science and a staff member in the Department of Electrical Engineering at Braude College of Engineering in Karmiel, Israel. She holds a Ph.D. in Education in Science and Technology, having conducted her doctoral research under the supervision of Orit Hazzan.