From memory to judgment, which skills really matter
For generations, school rewarded those who could remember. Dates, formulas, definitions, memory was the mark of competence. Today, in a world where any answer is available in seconds on any conversational assistant, that mark has stopped working. The unique value of being human no longer lies in having the answer, but in knowing which question to ask, how to judge what you receive and how to place it in a context the machine does not know.
It is one of the strongest convergences in the EduNext 2026 report by the Look4ward Observatory (Luiss and Intesa Sanpaolo), with interviews with leaders of companies, universities and institutions describing the same shift, phrased in different ways but with the same structure. The education system must stop privileging memory and start privileging judgment. It is not a new idea in pedagogy, but AI makes it urgent and no longer postponable.
Value shifts from memory to judgment

Where the value of human skills is shifting. Source EduNext 2026.
Subject content does not lose meaning, but it changes function. It is no longer information to store, it is a tool to train reasoning. History is not studied to know the dates, but to develop long-term pattern recognition. Mathematics is not studied to perform calculations, but to learn how to frame a problem correctly. It is a shift that one leader interviewed in the report sums up clearly.
"I graduated almost entirely thanks to memory. The effort, today, is to push people to develop judgment, because that is what they will need. The rest the machine does, better than us."
More information does not mean less ambiguity
There is a distinction that emerges forcefully from the interviews, and it is valuable for understanding where AI helps and where it does not. Artificial intelligence vastly expands informational power, namely the volume of data and scenarios we can access. But it does not reduce decisional ambiguity. If anything it amplifies it, because multiplying the alternatives does not help you choose among them. AI generates possibilities; it is up to the human to step in, weigh them and decide. This is why experience changes nature.
"Experience, which used to be memory, is no longer memory. It is the ability to resolve ambiguity, to define what is right and what is wrong in a context the machine neither knows nor can know."
The cognitive fork: oracle or tool

It is not access that creates the gap, but the quality of use. Source EduNext 2026.
Here is the point we care about most, because it concerns school and training directly. The spread of AI tools carries a paradox that the rhetoric of technological inclusion tends to hide. It is not access that creates the gap, but how the tool is used. Those who use AI as an oracle, delegating the answer without exercising judgment, lose cognitive skills through disuse. Those who use it as a tool, to ask better questions and evaluate the outputs, multiply their own abilities.
The fork is self-reinforcing. Those who start cognitively advantaged use AI more skilfully and widen the gap; those who start behind use it passively and further deplete their resources. Educational research has long documented that offloading memory to an external tool reduces retention (Sparrow and colleagues, 2011), and that those with fewer critical tools struggle to recognise their own limits (Kruger and Dunning, 1999). A leader interviewed puts it in an image that stays with you.
"In the digital world we have all the water we could want, but being able to tell drinkable water from muddy water is one of the most important skills. And this is not learned from AI, it is learned despite AI."
The 15% that counts
One of the most striking accounts in the report puts numbers on the question and reframes it decisively. In most tasks the machine already does better than us; what matters is the residual share, the one where human judgment is irreplaceable.
"In 85% of cases the machine's output is superior, faster and cheaper. What counts is that 15%, but you have to be very clear about which it is and invest deeply exactly there."
The training response, then, changes not in breadth but in nature. It is not about shielding a vast area of skills from substitution, but about identifying precisely where that 15% lies and going deep into it. It is a metacognitive skill, namely understanding what is certain, what is probable and what is improbable among the scenarios AI presents all with the same confidence. And it is exactly what education systems find hardest to develop, because it requires getting people used to living with uncertainty rather than eliminating it.
Bloom's problem and the promise of tutoring

The advantage of one-to-one tutoring measured by Benjamin Bloom, source Bloom 1984.
There is also good news. In a now-classic experiment, Benjamin Bloom (1984) showed that one-to-one tutoring produces learning outcomes about 98% higher than traditional classroom teaching, namely two standard deviations. The problem, ever since, has been cost, namely no education system can afford a dedicated tutor for every student. AI agents open, for the first time, the possibility of scaling that personalisation at accessible cost, answering each learner and giving the teacher a picture of the most recurring difficulties.
On one condition, though. The benefit materialises only if paired with the ability to critically assess what the agent returns. Without judgment, the tutor becomes an oracle again, and we are back at the fork. Technology, on its own, is never enough, because the teacher matters.
Transversal skills, what they are and why they weigh more now
In Italian schools, when people talk about judgment, the phrase that keeps coming up is competenze trasversali, transversal or soft skills. They are the abilities that do not belong to a single subject and that you carry into any context, such as learning to learn, working with others, communicating, organising your own work. The Council of the European Union Recommendation of 22 May 2018 places them among the eight key competences for lifelong learning, starting with the personal, social and learning to learn competence. Italy's 2019 guidelines for the pathways for transversal skills and guidance (ministerial decree 774, pathways renamed school-to-work training from the 2025/26 school year) set them as the educational goal of those pathways. And in 2025 a dedicated law arrived as well, law no. 22 of 19 February, on introducing the development of non-cognitive and transversal skills.
For years they were treated as a side dish to the subjects, useful but secondary. AI flips the hierarchy. Content the machine retrieves in a second loses value as information to own, while transversal skills are exactly what you need to ask the right question, judge what comes back and decide in a context the tool does not know. In other words, the judgment EduNext talks about is not one more subject to add to the curriculum. It is the transversal skill par excellence, and it is trained inside every subject, not next to it.
What it means for training
None of this is a theoretical abstraction, it is the heart of how we think about training. Teaching in the AI era does not mean transferring information the machine retrieves better than we do, but training people to ask better questions, evaluate outputs and recognise the 15% that stays human. It is training that puts critical thinking and judgment at the centre, from students to teachers, from administrative staff to those who lead an organisation.
To bring training that develops judgment and critical thinking to your school or organisation, you can get in touch here, discover the Retoria method, or explore how AI affects logical and linguistic skills and what the 5 preconditions for adopting AI are.
Methodological note
The (anonymised) leader quotes and the analysis of the metamorphosis of skills come from the EduNext 2026 report (chapter 2.3), based on thirty-two in-depth interviews conducted between December 2025 and April 2026 using Constructivist Grounded Theory. The one-to-one tutoring figure is Benjamin Bloom's "2-sigma problem" (1984). The references to reduced retention from offloading memory (Sparrow et al., 2011) and to the metacognitive gap (Kruger and Dunning, 1999) are drawn from the literature cited in the report. The chart percentages are rounded for illustration.
Sources
- Look4ward Observatory, Luiss Research Center for Strategic Change "Franco Fontana" & Intesa Sanpaolo (2026). EDUNext: New scenarios for Education and skills in the AI era (CC BY 4.0).
- Bloom, B. S. (1984). The 2 Sigma Problem. Educational Researcher, 13(6), 4-16.
- Sparrow, B., Liu, J. & Wegner, D. M. (2011). Google Effects on Memory. Science, 333(6043), 776-778.
- Kruger, J. & Dunning, D. (1999). Unskilled and Unaware of It. Journal of Personality and Social Psychology, 77(6).
- Council of the European Union (2018). Council Recommendation of 22 May 2018 on key competences for lifelong learning. Official Journal of the European Union, C 189.
- Italian Ministry of Education (2019). Guidelines for the pathways for transversal skills and guidance (in Italian), adopted by ministerial decree 774 of 4 September 2019.
- Italian law no. 22 of 19 February 2025 (in Italian), on introducing the development of non-cognitive and transversal skills.
- Italian Chamber of Deputies, briefing on decree-law 127 of 2025 (in Italian), converted by law no. 164 of 30 October 2025.
- VanLehn, K. (2011). The Relative Effectiveness of Human Tutoring and Intelligent Tutoring Systems. Educational Psychologist, 46(4).