Scaffolded Regulation or Internalized Autonomy? Comparing Two Instructional Designs for Agentic LLM Use in EFL Writing
作者
Hyein Chong
出处
期刊:Hyeondae yeong'eo yeongmunhag [Modern Studies in English] 日期:2026-05-31卷期号:70 (2): 161-189
标识
DOI:10.17754/mesk.70.2.8
摘要
Large language models have entered EFL writing classrooms faster than instructors anticipated, and how to teach with them effectively remains unsettled. While these tools offer immediate feedback, they also risk reducing the productive struggle essential for language development when students rely on them to generate ideas or structure. This study examines instructional conditions that promote agentic, self-directed AI use. Using a quasi-experimental mixed-methods design, it compares two semesters of the same EFL writing course at a Korean national university. Group 25-1 received detailed scaffolding with a metacognitive prompt, while Group 25-2 was given a single guiding principle. Both conditions limited passive AI use but through different mechanisms. Group 25-1 showed stronger in-task regulation and engagement, whereas Group 25-2 demonstrated greater critical agency and text development. These results point to distinct external and internal pathways to agentic AI use, each shaping different aspects of learner agency, though their longer-term developmental effects remain unclear. (Chungnam National University)