With generative and agentic AI, adaptive learning finally becomes concrete: diagnosis, large-scale personalization and inclusion for the success of all students.
Artificial intelligence now occupies a central place in debates about the future of higher education. However, one of its most promising applications has long remained one of the least concrete: adaptive learning, which mobilizes AI and learning data to adjust in real time the level of difficulty, pace and resources offered to each student. What yesterday was an educational intention is today, thanks to rapid progress in generative and agentic AI, an operational reality.
An old promise, long remained theoretical
Adaptive learning is not new: its roots go back to the first intelligent tutorial systems of the 1970s, before accelerating with the digital education of the 2000s. Unlike traditional systems, based on standardization of content, it is based on the opposite principle: each learner progresses differently, and it is up to the system to adapt to the individual. Well designed, this personalization promotes engagement, improves knowledge retention and optimizes journey efficiency (Pane, Steiner, Baird, & Hamilton, 2015). But until recently, this promise remained largely theoretical: the algorithms existed, as did the educational intentions, without the two really meeting on a large scale. The United States and the Nordic countries entered this field earlier; French higher education still remains behind.
What generative and agentic AI really changes
The main obstacle to adaptive learning has long been the paucity of available data. Learning management platforms were limited to traditional indicators, connection times, grades, access to resources, without capturing the cognitive strategies or affective states of students. Deprived of this granularity, the algorithms lacked benchmarks to calibrate their adjustments, and their effects remained very dependent on the conditions of educational implementation (Holmes, Bialik, & Fadel, 2019). The arrival of generative models, capable of understanding natural language, dialoguing and producing content on demand, is already changing the situation: it is becoming possible to interact in detail with each student and understand where they are really stuck. But it is the agentic dimension that takes an additional step: beyond the simple generation of content, these systems can now orchestrate a complete educational sequence themselves, chain together diagnosis, exercise, correction and remediation without human intervention at each stage, mobilize several tools independently and adjust their strategy continuously as progress is observed. Adaptive learning thus ceases to be a simple recommendation engine to become a true educational co-pilot, capable of acting and not just suggesting.
A concrete lever for success and inclusion
It is in this now tangible ability to adjust to each profile that the most decisive challenge lies: reconciling academic selectivity and equality of opportunity. Far from standardizing courses, a school can deploy diagnostic tools upon admission that precisely identify the acquired knowledge and shortcomings of each candidate, to offer truly personalized refreshers rather than a uniform common core. Students can now benefit from intelligent scientific tutoring capable of reformulating a poorly understood concept as many times as necessary, augmented learning environments that make abstract concepts manipulable, and a personal assistant to help them distribute their workload. It is an excellent way to prevent dropping out, by detecting weak signals before the break, but also to fight against social self-censorship, by revealing to each student a potential that their original environment could have hidden.
A strategic choice now essential
The importance of adaptive learning in transforming the educational world remains underestimated, even though the technical conditions for its success through AI have finally been met. Individualizing the educational experience is no longer an unattainable technical feat: it responds to an expectation of student autonomy, but also to a requirement for social justice too often relegated to the background of discussions on academic excellence. Adaptive technology, however effective it may be, will only produce lasting effects if it is linked to the educational, organizational and evaluative dimensions of the institutions that deploy it. It is not enough to install a tool: we must also rethink teaching practices and support the teams, otherwise the technology remains a superficial layer.
Establishments can therefore no longer be content with observing these developments from a distance. Adaptive learning must become a strategic axis in its own right, articulated with ambitious data governance and a clear educational vision. It is on this condition that personalization, long promised, will finally become a real lever for success and inclusion, capable of supporting each student in their uniqueness while maintaining a common academic requirement.
References
Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial intelligence in education: Promises and implications for teaching and learning. Center for Curriculum Redesign.
Pane, JF, Steiner, ED, Baird, MD, & Hamilton, LS (2015). Continued progress: Promising evidence on personalized learning. RAND Corporation.
