Lepida, in collaboration with the University of Bologna, presented a study originating within ER2Digit at EDUCON2026, a flagship conference of the IEEE Education Society and a premier venue for STEM education.

The focal point generating significant interest was the study's cohort and context: 20 final-year master's degree students who undertook a foundational literacy experiment.

The resulting methodology, colloquially termed "learning by stumbling," transformed AI errors and imperfect responses into opportunities for deep comprehension.

The introduction of mandatory paper-based assessments, combined with the rotation of roles in drafting, validating, and documenting interactions with Large Language Models (LLMs), generated, through deliberate "friction", a level of critical reflection far deeper than that produced by frictionless interactions.

Four key insights emerged; challenging uncritical trust in AI; rigorously documenting one's work; subjecting outcomes to structured critical review; safeguarding ethical standards and data protection protocols.

While the research presented certain limitations, its core message resonated: integrating AI into unprepared environments demands significantly more time, mediation, and awareness than is commonly assumed, proving that "friction" can serve as a vital educational resource.