For decades, universities have relied on graduate students and postdoctoral scholars to power the research enterprise. The arrangement has produced extraordinary scientific advances, but it has also created a persistent imbalance. While estimates vary, tens of thousands of postdoctoral scholars compete each year for only a few thousand tenure-track openings. The exact figures vary by field and year, but the underlying reality is unmistakable: The academic system produces far more aspiring researchers than it can employ in stable academic roles.
The consequences are well-known. Many researchers spend years moving through temporary appointments with no clear path to a permanent academic position. While postdoctoral training provides valuable opportunities for intellectual growth and research independence, only a minority of postdocs ultimately secure tenure-track jobs.
The benefits of postdoctoral training are real. Postdocs gain intellectual independence, deepen their expertise and play a central role in advancing research. They are often the engine of productivity in academic labs, contributing ideas, mentoring students and sustaining complex projects.
But the downside emerges at the point of transition. I recently participated in a Stanford University faculty search that attracted more than 500 applicants for a single position. The experience made visible what national statistics often obscure: The problem is not a shortage of talented researchers. It is a shortage of academic destinations for them.
The usual response to this imbalance is to call for more faculty positions. While desirable, this is unlikely to occur at the scale required and may even move in the opposite direction as institutions adopt new technologies. If demand cannot expand significantly, then supply must be adjusted.
A new variable is now entering the equation: agentic AI. Unlike conventional software tools, agentic systems can pursue multistep research goals: monitoring literature, generating hypotheses, analyzing data and refining results with limited human supervision.
The key point is not that agentic AI replaces human researchers. It does not. But it can reduce the number of human trainees required to carry out certain kinds of work.
Agentic AI systems are particularly strong at scale and continuity. They can track vast bodies of literature in real time, maintain structured maps of evolving fields and explore multiple lines of inquiry simultaneously. Tasks that once consumed large amounts of human effort, such as searching, organizing and synthesizing, can increasingly be handled by these systems.
This shifts the role of the human researcher. Instead of spending most of their time managing information, postdocs and Ph.D. students can focus more on defining meaningful questions, interpreting results and exercising judgment. In other words, the human contribution becomes more central, not less, but also more selective.
This creates a fundamental choice about how the academic research enterprise evolves. The critical question is not whether universities adopt agentic AI. They almost certainly will. The question is how they will use it.
One path is what might be called the expansion model. In this model, AI increases productivity while institutions continue to train the same or even higher number of graduate students and postdocs. Research output rises, but the workforce imbalance remains.
The alternative is the substitution model. In this model some tasks currently performed by trainees are delegated to AI systems. Research productivity is maintained or increased, but fewer postdocs are required. Over time, the size of the academic pipeline begins to align more closely with the number of available faculty positions.
Reducing dependence on postdoctoral labor does not necessarily mean reducing doctoral education. It means fewer Ph.D.s deciding to go on and do a postdoc.
The distinction matters. The expansion model increases research output while leaving the workforce imbalance largely unchanged. The substitution model offers a path toward gradually reducing that imbalance.
Whether substitution occurs will depend less on technology than on incentives. Principal investigators are rewarded for productivity, often measured in publications and grants, not for efficiency or workforce sustainability. If adding more trainees remains the easiest way to increase output, the system will continue to expand.
Changing this dynamic would require coordinated shifts:
Funding agencies could reward proposals that achieve strong results with smaller teams.Universities could place greater emphasis on mentorship quality rather than trainee quantity.Departments could track and share career outcomes more transparently.Laboratories could reinvest savings from reduced personnel costs into better support for fewer researchers.
While the expansion model is likely to be the one adopted by most research groups, if even a fraction of such groups moves in the substitution direction, the effect could be significant. The number of postdocs competing for faculty roles could decline, and those who remain in the pipeline might be better positioned for success.
This would also reshape the role of the postdoc itself. Rather than serving as a broad holding category, it could become a more focused stage of preparation for research leadership. Fewer individuals would pass through it, but those who do would be closer to faculty readiness.
At the same time, the system would need to expand and legitimize alternative career paths. Not all Ph.D. graduates need to pursue postdoctoral training, particularly when their long-term goals lie outside research-intensive academia. Strengthening connections to industry, government, teaching-focused institutions and entrepreneurship would provide more viable and respected options.
Taken together, these changes could transform what is currently experienced as a crisis into something more manageable: a system that produces fewer, better-supported trainees, aligned more closely with available opportunities.
None of this diminishes the appeal of an academic career for those who are deeply committed to it. For the right individuals, it remains a uniquely rewarding path. But it does suggest that the system supporting that path needs to evolve.
The postdoc bottleneck is often treated as an unavoidable feature of academic life. It is not. It is the result of choices about how research is organized, funded and staffed.
Agentic AI technology is arriving regardless. The real question now is whether academic institutions will use it to produce more researchers than the system can absorb, or to create a research workforce better aligned with the opportunities available.
Richard M. Reis is research liaison with the Quantum Mechanical Engineering Laboratory at Stanford University and author of Tomorrow’s Professor: Preparing for Academic Careers in Science and Engineering (Wiley-IEEE Press, 1997). He can be reached at reis@stanford.edu.