From Generative AI to Agentic AI in Higher Education: A Behavioural Experiment on Delegation Using Diffusion of Innovations
Education Sciences · 12.11.2026 · 16 (11) · 1941
DOI: 10.3390/educsci16111941 ↗Kurz gefasst
Fachartikel in der Zeitschrift Education Sciences, online erschienen am 12.11.2026. Verfasst von Raghu Raman und 2 weiteren Autor:innen. Die Arbeit nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science and Technology Council of Taiwan, Ministry of Education.
Abstract (Originalsprache)
In higher education, agentic AI goes beyond drafting content to creating learning management system (LMS) artefacts and scheduling releases under teacher oversight. Adoption research has modelled intention towards tools that draft rather than act, leaving delegation to a system that acts unobserved. This study estimated the effect of assignment to an action-taking scenario on teachers’ workflow choices, using Rogers’ Diffusion of Innovations (DOI) theory to explain delegation within that scenario. We randomly assigned 290 university teachers to a GenAI drafting-only condition (n = 140) or an LMS-integrated agentic condition (n = 150); each made step-by-step choices in a scenario-based six-step quiz workflow combining content drafting with student-facing platform actions. Teachers assigned to the agentic scenario chose AI support for more steps (d = 0.65), but the gain sat almost entirely on content drafting (d = 0.99) and was negligible for platform actions (d = 0.12). This content–platform boundary was sharp: 92% chose to delegate question writing, while 77% chose to schedule releases manually. Most chose limited autonomy (48% selected draft-only, 35% selected approval-gated publishing, 5% selected auto-publish). The five DOI attributes and controls explained substantial variance in an Agentic Use Behaviour Index (AUBI) of delegation, autonomy, and permissions (R2 = 0.65). Complexity, understood as oversight burden, was negatively associated with AUBI, and its association with non-manual engagement was stronger in the agentic scenario; observability was not significant. The direction of the AUBI results held across four weightings and count and ordinal models, and exploratory clustering suggested four adopter profiles forming a permission ladder. Extending DOI from tool adoption to task delegation, the results favour draft-first defaults, approval gates, and staged permissions, with field replication using LMS logs a priority.
Förderung
- National Science and Technology Council of Taiwan
- Ministry of Education
Zitieren
APA
Raman, R., Tang, K., & Nedungadi, P. (2026). From Generative AI to Agentic AI in Higher Education: A Behavioural Experiment on Delegation Using Diffusion of Innovations. Education Sciences, 16(11), 1941. https://doi.org/10.3390/educsci16111941
Vancouver
Raman R, Tang K, Nedungadi P. From Generative AI to Agentic AI in Higher Education: A Behavioural Experiment on Delegation Using Diffusion of Innovations. Education Sciences. 2026;16(11):1941. doi:10.3390/educsci16111941
BibTeX
@article{Raman2026,
author = {Raman, Raghu and Tang, Kai-Yu and Nedungadi, Prema},
title = {From Generative AI to Agentic AI in Higher Education: A Behavioural Experiment on Delegation Using Diffusion of Innovations},
journal = {Education Sciences},
year = {2026},
volume = {16},
number = {11},
pages = {1941},
doi = {10.3390/educsci16111941}
}
EUVIMED