A student is asked to evaluate a vaccination strategy. Using generative AI, the student produces a clear, well-structured analysis grounded in global evidence. The recommendation is logical, comprehensive,and wrong. It overlooks unreliable cold-chain infrastructure, intermittent electricity, and low public trust. Under these conditions, the strategy would almost certainly fail.
This is not a failure of analysis. It is a failure of contextual judgment: the ability to determine what matters, under what circumstances, and how knowledge should be applied in a particular setting. Facts, analysis, and even sound reasoning are necessary for good decisions, but they are rarely sufficient. Effective judgment depends on recognizing which assumptions no longer hold, which constraints are decisive, and how people and institutions are likely to respond.
Steve Jobs famously described the computer as a “bicycle for the mind,” a tool that amplified human capability by making us dramatically more efficient thinkers.
Historically, technologies that automate work have done more than increase productivity. They have also changed how expertise develops by altering the work through which people learn.
Generative AI represents a different kind of technology. It does not simply help us think faster; it increasingly shapes the analyses we produce, the questions we pursue, and the options we consider. It doesn’t just pedal. It steers.
AI is changing how expertise itself is developed.
We have seen versions of this dynamic throughout our careers in technology, university administration and diplomacy. For example, product decisions could be technically correct but analyses failed because they overlooked user behavior; universities where carefully designed strategies faltered because they underestimated institutional culture; and diplomacy where analytically rigorous positions proved ineffective because they failed to account for political realities. In each dynamic, success depended less on producing technically correct answers than on exercising sound judgment about context.
As AI lowers the cost of producing analysis, the central constraint organizations face is shifting. Analysis is becoming abundant. The ability to exercise informed judgment is becoming comparatively scarce.
The AI Talent Paradox
AI is changing how expertise itself is developed.
For decades, most organizations relied on an implicit production system for expertise. Entry-level professionals did more than complete assigned tasks. They learned by performing them. They tested assumptions against reality, received feedback from more experienced colleagues, observed how decisions unfolded in practice, and gradually accumulated the pattern recognition that distinguishes expertise from technical competence.
When work changes, learning changes. And when learning changes, so too does the production of judgment.
Productive work and professional development were inseparable. Organizations were not simply generating output; they were simultaneously producing the next generation of judgment.
Generative AI is beginning to uncouple those two functions. Many analytical tasks that once required days of junior effort can now be completed in minutes. That transformation creates undeniable gains in efficiency. Yet it also removes many of the experiences through which expertise has traditionally developed.
When work changes, learning changes. And when learning changes, so too does the production of judgment.
Generative AI can now summarize vast bodies of information, synthesize established knowledge, generate persuasive arguments, and perform many analytical tasks that until recently were the foundation of human engagement. In a growing number of occupations, producing competent analysis is no longer the defining human contribution. The scarcity is shifting elsewhere.
To understand that shift, it is useful to distinguish four related but different capabilities.
Information consists of facts and observations. Analysis organizes information into explanations and recommendations. Expertise develops through accumulated experience applying that analysis under real-world conditions. Judgment determines which analysis should ultimately guide action.
Generative AI increasingly excels at the first two. Humans remain responsible for the latter two.
What Aristotle Taught Us
More than two millennia ago, Aristotle described the capability to apply knowledge as “phronesis,” or practical wisdom, the ability to make sound decisions where general rules alone are insufficient.
Unlike “episteme,” or theoretical understanding, and “techne,” or technical skill, practical wisdom develops through experience.
Practical wisdom emerges through repeated encounters with uncertainty, feedback, and consequence. It cannot simply be transferred through instruction or acquired by reading the correct answer.
We refer to this higher-order capability as contextual judgment, which we characterize as the ability to determine what matters, for whom, and under what circumstances.
Contextual judgment requires recognizing which assumptions deserve scrutiny, identifying information that may be missing, adapting general knowledge to changing conditions, and understanding the institutional, political, technical, and human contexts in which decisions unfold.
Context is not simply background information. It is often the deciding factor separating technically correct recommendations from successful outcomes. Organizational culture, stakeholder incentives, public trust, implementation constraints, and political realities frequently determine whether analytically sound recommendations succeed or fail.
These considerations are not peripheral to good decisions; they are often their essence. That distinction becomes increasingly important as AI improves.
Recent research illustrates both the promise and the limits of this transformation. OpenAI policy researcher Tyna Eloundou and colleagues estimate that approximately 80% of the workforce in the United States could see at least 10% of their work tasks affected by large language models. Although, the study measures task exposure rather than job displacement.
As AI assumes a growing share of analytical work, the relative value of producing analysis declines while the value of evaluating, adapting, and applying that analysis grows.
At its core, this is more than a story about artificial intelligence. It is a story about institutions and how they transform novices into experts.
Evidence from AI’s use in clinical medicine underscores the same point. A June 2026 Nature Medicine benchmarking study found that specialized clinical AI tools performed no better than general search assistance on real physician queries, despite strong performance on standardized medical exams.
The Stanford-Harvard ARISE network’s inaugural State of Clinical AI report reaches a similar conclusion. Across multiple recent evaluations, systems that perform exceptionally well on benchmark tests often encounter greater difficulty when deployed in real clinical environments characterized by incomplete information, changing conditions, workflow constraints, and human variability. High analytical performance does not automatically translate into sound decisions.
The central challenge increasingly becomes recognizing where those recommendations may fail rather than the more automated task of generating the recommendations in the first place.
The emerging scarcity focuses on the expertise required to bridge the gap between analytical correctness and real-world relevance. This is a significant change from a system where intelligence and information often limited progress.
As contextual judgement becomes more important, so does the question of how people develop it.
Intergenerational Expertise Matters
The AI Talent Paradox is not simply that AI changes work. Rather, AI changes the process through which organizations reproduce expertise across generations.
The consequences unfold gradually rather than suddenly. Expertise is rarely acquired through isolated moments of insight. It develops through repeated encounters with uncertainty, feedback, disagreement, and consequence.
We argue that the AI Talent Paradox arises from three reinforcing dynamics.
First, generative AI dramatically reduces the cost of producing high-quality analysis. Second, organizations naturally redesign work by automating many of the entry-level analytical tasks that have traditionally served as apprenticeships for developing professional expertise. Third, as those formative experiences diminish, the future supply of contextual judgment contracts precisely when organizations need it most.
The paradox is that the same technology that is increasing the value of human judgment may also be weakening the system through which that judgment has historically been produced.
At its core, this is more than a story about artificial intelligence. It is a story about institutions and how they transform novices into experts.
AI increasingly produces analysis. Organizations must become more intentional about producing expertise.
Emerging labor-market evidence suggests that this transition may already be underway. Using payroll records covering millions of U.S. workers, Brynjolfsson, Chandar, and Chen report disproportionate declines in employment among workers aged 22 to 25 in occupations with high exposure to generative AI. Employment among more experienced workers remained comparatively stable. The authors caution against interpreting these findings as definitive causal evidence and acknowledge that broader labor-market dynamics contributed to early employment trends.
The findings raise an important question for employers. As AI performs more of the analytical work traditionally assigned to early-career professionals, how will those professionals acquire the experience through which expertise develops?
SignalFire’s 2026 State of Tech Talent Report, for example, finds that engineering hiring has proven considerably more resilient than many other early-career knowledge roles as AI expands the scope of work organizations are able to undertake.
Rather than focusing on the issue of whether AI eliminates jobs, the central question should be: Which kinds of work does AI change? If automation removes activities that historically functioned as apprenticeships for professional judgment, organizations risk weakening the developmental pipeline where future experts are formed.
As contextual judgment becomes more valuable, organizations will have stronger incentives to recruit and retain experienced professionals and to invest more deliberately in developing them. Yet market incentives alone are unlikely to solve the problem. Much of the expertise developed within one organization is transferable to others.
But much of that expertise moves with the people who acquire it. An organization that invests in developing someone may not be the one that ultimately benefits from that investment. The long-term scarcity, therefore, is not intelligence, information, or even analytical capability. It is experienced judgment.
AI increasingly produces analysis. Organizations must become more intentional about producing expertise.
Redesigning the Production of Judgment
Institutions must continue to produce the contextual judgment given that AI cannot. Redesigning apprenticeship does not mean preserving inefficient work for its educational value or slowing the adoption of AI where it adds clear value. It does mean finding new ways for professionals to gain the experience that analytical work once provided.
Organizations will need to create deliberate opportunities for early-career professionals to frame problems, evaluate AI-generated analyses, defend their reasoning, receive structured feedback, and observe how decisions succeed — or fail — in practice.
AI should accelerate learning, not eliminate it.
AI will give the next generation of professionals analytical capabilities that previous generations could scarcely have imagined. Institutions now must make sure they continue to develop the experience and judgment needed to use those capabilities well.
Mid-career professionals will also occupy an increasingly pivotal role. They represent the bridge between expertise developed largely before the widespread adoption of generative AI and a workforce learning alongside it. Their task is no longer simply to supervise work.
They will need to cultivate judgment by helping colleagues understand not only how to use AI effectively, when to question it, and where its limitations lie. They will also need to train junior colleagues to see how context can reshape apparently straightforward recommendations. Mentorship becomes less a transfer of knowledge than a process of developing professional judgment.
Universities have an important role as well, building on their strengths in preparing students to evaluate assumptions, recognize contextual limitations, and make decisions under uncertainty.
Technological revolutions change both what societies produce and what they must become good at producing. As analysis becomes increasingly abundant, expertise formation can no longer be treated as an incidental byproduct of this process. It becomes a deliberate institutional responsibility.
AI will give the next generation of professionals analytical capabilities that previous generations could scarcely have imagined. Institutions now must make sure they continue to develop the experience and judgment needed to use those capabilities well.
* The views expressed in this article are those of the author and do not necessarily represent the views of their home institutions.
