Becoming an AI-Empowered Agentic Learner: A Multiple-Case Study of Learner Collaboration with Generative AI in Higher Education
Year: 2026
Type: article
Abstract: The rapid development of generative artificial intelligence (GenAI) is reshaping learning in higher education. However, its presence does not automatically lead to meaningful or high-quality learning. While prior research has explored how GenAI can support learning through tool design, instructional interventions, and analyses of general usage patterns, less is known about how learners who collaborate effectively with GenAI actually think, act, and maintain control over the learning process in real situations. To address this gap, this study focuses on the AI-empowered agentic learner, who retains ownership of learning goals and meaning-making while actively organizing, coordinating, and evaluating human–AI collaboration and using GenAI to expand cognitive and practical capacities. This study adopts a qualitative multiple-case design to examine the key characteristics of AI-empowered agentic learners and how these characteristics are manifested across different stages of learning or problem solving. Using purposeful sampling, three higher education learners with high self-directedness and strong perceived GenAI use efficacy were recruited. The dataset consisted of 12 authentic human–AI collaboration cases. Two sources of data were collected: chat logs with GenAI tools and semi-structured retrospective interviews conducted around each selected case. For analysis, agency was used as the overarching concept, while self- regulated learning (SRL), self-directed learning (SDL), and active learning (AL) served as complementary analytical lenses. A hybrid deductive–inductive thematic analysis was conducted to identify both theoretically aligned and context-specific manifestations of learner agency in human–AI collaboration. The findings show that AI-empowered agentic learners are characterized by three interrelated tendencies. First, they actively construct human–AI collaboration before and during task engagement by diagnosing needs, defining goals, mobilizing resources, and planning collaborative strategies. Second, they maintain cognitive ownership and quality control throughout the collaboration process by monitoring task progress, evaluating their own understanding and expression, and critically examining the credibility, usefulness, and limitations of GenAI outputs. They also use GenAI not merely for surface-level assistance but to support higher-order thinking such as analysis, synthesis, evaluation, and creation. Third, they demonstrate selective uptake and adaptive adjustment by filtering, revising, and reorganizing GenAI-generated suggestions in light of task demands, situational constraints, and quality expectations, and by adjusting prompts, interaction strategies, or even tools when collaboration becomes ineffective. These characteristics unfold processually across the forethought, performance, and self-reflection phases of learning. This study contributes to current human–AI collaboration research by shifting attention from how GenAI supports learning to how learners sustain and enact agency in human–AI collaboration. It also extends understandings of the agentic learner from an individual self-regulating subject to an active designer and manager of a human–AI collaborative system. Practically, the findings offer a more concrete reference for identifying and fostering productive human–AI collaboration in higher education.
Source: Työväentutkimus Vuosikirja
Author: Yang, Danyu
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
Keywords: generative AI, human-AI collaboration, learner agency, agentic learning, student self-regulation, self-directed learning, active learning, higher education, self-direction, higher-order thinking skills, prompt adaptation, critical evaluation of AI outputs, self-reflection, chat logs, metacognition
SDG: Quality education
Open Access status: green
OpenAlex ID: W7198694503
