Producing a correct answer no longer proves that a student understands: generative AI can do it in seconds. Learning Curves, published by Wiley in September 2026, shows that educational systems are adapting unevenly in how they evaluate students in the face of this reality. Juan M. Lavista Ferres and Frank Nagle have assembled data from several countries to establish that the diploma crisis is manifesting unequally depending on context, and that economies with fragile foundational skills will suffer significant effects.

The essentials

  • Generative AI changes the scope of certifications based on producing correct answers, which tends to accelerate diploma devaluation in countries where foundational learning is already weak.
  • The 432-page book compares the effects of AI on school education, short-term training, professional certification, and entry-level careers across several countries.
  • ECLAC documents strong educational inequalities in Latin America and the Caribbean: its evaluation systems often rely on content reproduction, precisely when AI is transforming the difficulty of that reproduction.
  • Two trajectories are opening up: possible exam reform centered on reasoning in real situations, or potential devaluation that could worsen existing inequalities.

Two researchers, one shared diagnosis

Juan M. Lavista Ferres heads the AI for Health Institute at Microsoft. Frank Nagle is a professor at Harvard Business School, specializing in the economics of open-source software and measuring the effects of technology on productivity. Their collaboration on Learning Curves brings together two perspectives that discussions about AI in education usually keep separate: the technical capabilities of the tools, on one hand, and the economic and institutional structures that determine whether these tools serve or disrupt education, on the other.

The book is a comparative inquiry into what happens when a content-production technology enters systems whose only quality signal is content production. The authors diagnose without prescribing. The diagnosis is severe for systems that have changed little since writing an essay or solving a problem at home constituted proof of effort and understanding.

An old fragility made visible and urgent by AI

The core of Learning Curves rests on a simple idea to state, difficult to accept institutionally: generative AI does not create a new problem in education. It makes an old one visible and urgent.

For decades, part of certification has rested on an implicit assumption: if a student produces the right answer, they probably understood. This assumption was already questionable before AI; studies on cramming and short-term memorization have shown this since the 1980s. But it was tolerable, because producing a correct answer required at least minimal effort. Generative AI greatly reduces that effort. It significantly weakens the validity of the assumption.

The collective work examines differences in AI diffusion across educational and economic contexts. Systems that combine oral exams, practical situations, and continuous evaluations based on demonstration show greater robustness. Systems where the final evaluation is a written reproduction exam show structural vulnerabilities. And in this second category, several Latin American economies figure according to the authors.

ECLAC published in June 2026 Educación, desigualdad y desarrollo social inclusivo. The report analyzes inequalities in coverage and learning and notably recommends strengthening teaching competencies and capacities, precisely because foundational skills remain unevenly distributed there. According to the authors, when a student has weak reading and formal reasoning skills, access to a tool that produces fluent texts can mask gaps without filling them. The certification that follows this masking tells an employer less useful information.

Entry-level employment under pressure

Learning Curves does not stop at the classroom. One of the most concrete contributions of the book is its treatment of American community colleges and their equivalents in other countries, these two-year short-term training programs that constitute, for millions of young people, the passage to the qualified job market.

These programs have long held their value through their proximity to employers. They certified verifiable practical skills, in sectors—health, logistics, basic accounting, customer service—where demand was stable. The book examines possible effects of AI on entry-level employment and emphasizes that aggregate effects remain uncertain. According to the book, report writing, structured data entry, responses to standardized requests do not disappear but evolve. They tend to demand less production and more supervision, validation, and correction of automated judgment.

Community college certifications have not evolved at the same pace. Several continue to attest to the ability to produce what AI can now produce. The employer finds itself facing a signal that turns out to be ambiguous: this diploma may indicate that the candidate understands what they are doing, or that they knew how to use available tools during their training.

This ambiguity also affects entry-level jobs in large technology companies, a direct research field for Nagle. According to the book, some recruiters report a shift in hiring criteria for junior positions: less attention to portfolio deliverables, more to technical interviews in situations without tool access, and to references that testify to autonomous reasoning. The market tends to seek its own replacement signals, without waiting for institutions to provide them.

Susskind, Acemoglu: two readings of one choice

The thesis of Learning Curves resonates directly with a theoretical debate that economist Daniel Susskind posed in A World Without Work: technology is not neutral in its direction. It can enhance human capacities or replace them, and this choice is a matter of institutional decision, not inevitable technical logic. Applied to school, this framework is illuminating: AI can either serve to evaluate what a student truly understands, or replace the cognitive effort that allowed it to be measured. The decision belongs to institutions, based on the priorities they set.

Lavista Ferres and Nagle fit into this logic without citing it explicitly. Their argument is that schools have the capacity to direct AI toward evaluating reasoning, but many of them, through inertia, let the tool fit into formats designed for another time.

Daron Acemoglu and Simon Johnson, in their analysis of technology and power, add a tension that Learning Curves treats less directly: who captures productivity gains when tasks reconfigure. In education, this question takes a precise form. If large universities and large companies adapt quickly, by revising their evaluation and hiring processes, while community colleges and vocational high schools in emerging economies lack the means to do so, AI reconfigures inequalities without reducing them. Well-resourced institutions access new competence signals; others risk losing the value of old ones without having new ones available.

Learning Curves is both more and less ambitious than a work of educational policy. More ambitious, because it compares very different national systems and documents this asymmetry rigorously. Less ambitious, because the authors remain cautious about policy recommendations and diagnose better than they prescribe.

The decisions that remain with institutions

The forward-looking question that the book leaves open is central: educational systems still have the time needed to choose their trajectory.

Two scenarios emerge from Learning Curves’ comparisons and ECLAC data. In the first, systems that reform their exams to evaluate reasoning in real situations, without tool access, in oral or practical formats, could rebuild a credible signal. The diploma would once again attest to a verified capacity, distinct from what AI produces. In the second, systems that maintain reproduction exams risk seeing their certification lose relevance. Some employers develop their own filters.

Candidates from countries with fragile foundational skills risk losing a key formalized access to the qualified job market.

These two scenarios are not symmetrical. The first requires active reform, expensive in teacher training and curriculum revision, precisely the type of investment that French universities already struggle to fund in a constrained budgetary context. The second could happen by default, without visible decision, through accumulation of micro-adaptations that bypass certifications without reforming them.

The signals to watch are concrete. The share of national exams integrating reasoning tests without access to AI tools, by country and level, will say within two or three years which trajectory systems have actually chosen. The evolution of entry-level employment rates for short-term training graduates in countries covered by the book will say whether the market has outpaced or followed institutions.

ECLAC underscores a collective need that Learning Curves formulates without exhausting it: ECLAC recommends targeted interventions for disadvantaged groups and development of cognitive, socio-emotional, and digital skills, before certification devaluation accelerates. These systems exist in some Scandinavian countries in the form of data-driven continuous tutoring systems. Their extension to low-resource contexts is a question of public investment, not available technology.

The angles the book does not explore

Two limits deserve mention. The first: Learning Curves is written from a North American and Anglo-Saxon perspective, even when it looks at Latin America. The evaluation systems it takes as implicit references for good practice—technical interviews, documented portfolios, continuous evaluation—are formats that assume small class sizes, teachers trained in formative assessment, and institutions with the means to implement them. In a high school in São Paulo or Mexico City with 40 students per class, direct transposition is illusory.

The second: the book treats the question of teachers themselves little. If certifications must evolve, teachers must be able to evaluate differently. Yet ongoing teacher training is precisely the blocking point that several OECD education systems have not resolved, independent of AI. Lavista Ferres and Nagle note the difficulty without dwelling on it. It is a blind spot that weakens the operational scope of their diagnosis.

Reasons to read this book

Learning Curves is one of the rare books on AI in education that moves past two symmetrical positions: that of the enthusiast who sees in AI a universal tutor, and that of the critic who sees only a threat to critical thinking. Lavista Ferres and Nagle start from data, compare systems, and pose an uncomfortable question for school administrators: if an evaluation can be passed by a machine, its value as a certificate of understanding weakens.

For a reader interested in comparative education, labor economics, or public policy in Latin America, the book is a serious entry point into a debate that institutions have been slow to formalize. It does not give a definitive answer on what schools should transmit in the age of intelligent machines. But it makes it harder to continue avoiding the question.


Bibliographic information Learning Curves: Rethinking Education from Classrooms to Careers in an AI-driven World Juan M. Lavista Ferres and Frank Nagle (with contributors) Wiley, September 2026, 432 pages


Sources

  1. Juan M. Lavista Ferres and Frank Nagle, Learning Curves: Rethinking Education from Classrooms to Careers in an AI-driven World, Wiley, September 2026, https://www.wiley.com/en-sg/learning-curves-rethinking-education-from-classrooms-to-careers-in-an-ai-driven-world-p-9781394452477
  2. ECLAC, Education, inequality and inclusive social development, August 2026, United Nations Economic Commission for Latin America and the Caribbean
  3. Daniel Susskind, A World Without Work: Technology, Automation, and How We Should Respond, Macmillan, 2020, https://us.macmillan.com/books/9781250243089/aworldwithoutwork/
  4. Daron Acemoglu and Simon Johnson, Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity, PublicAffairs, 2023