Turning 500+ Students into Teachers: A Semester-Long Study of an AI Teachable Agent in an Undergraduate Algorithms Course

Synthesis: Wang et al. (2026) present Explique, a platform integrating an AI teachable agent (Algorithm Apprentice) into an undergraduate algorithms course to operationalize learning-by-teaching (LBT) at scale. In an 11-week field deployment with 546 students and 3,809 student–agent LBT dialogues, they find that explanation-oriented dialogue behaviors (elaboration, showing reasoning) are associated with fewer incorrect quiz submissions, while external-content reuse is associated with more repeated attempts. The LBT condition corresponded to a modest reduction in expected quiz attempts versus a baseline reading activity.

The Need for Large-Scale LBT Evidence

LLM tools give students rapid solutions but may reduce opportunities for productive struggle and explanation generation that support conceptual learning. Learning-by-teaching offers an alternative by positioning students as tutors — yet evidence for LLM-based teachable agents remained limited, especially for longitudinal deployments and large-scale evaluations connecting LBT interactions to conceptual understanding in authentic courses.

Study Design

• Explique platform integrates the Algorithm Apprentice teachable agent into an undergraduate algorithms course.
• 11-week field deployment in a real course with 546 students.
• 3,809 student–agent LBT dialogues analyzed alongside quiz and survey data.
• Generalized linear mixed-effects models linked dialogue behaviors to learning outcomes.

Key Findings

• Explanation-oriented dialogue predicts success. Dialogue behaviors such as elaboration and showing reasoning were associated with fewer quiz attempts (fewer incorrect submissions).
• External-content reuse hurts. Direct reuse of externally sourced content was associated with slightly more repeated attempts (more incorrect submissions).
• Modest LBT benefit. Compared to a baseline reading activity, the LBT condition corresponded to a modest reduction in expected quiz attempts, though confounded by substantial differences in time-on-task.
• Engagement varies in depth/authenticity. Students engaged consistently in multi-turn teaching interactions over the semester, though depth and authenticity varied, including instances of direct reuse of external content.

Implications

1. LBT with LLM teachable agents scales to authentic, large-enrollment courses.
2. Encourage explanation, discourage content-dumping — the quality of teaching interaction (elaboration vs. reuse) predicts learning, guiding system and instruction design.
3. Connect LBT behavior to conceptual understanding at scale is feasible with platform-based measurement.

Connected Concepts

• Learning By Teaching
• Generative AI
• CS Education
• Desirable Difficulties
• Scaffolding
• Higher Ed

Connected Articles

• Chatgpt Teachable Agent Programming Lbt 2024 — ChatGPT as a teachable agent in programming
• Prompting Teachability Novice Personas Lbt 2026 — Designing novice personas for teachability
• AI Tutor Safety Harms — Safety and harms of AI tutoring
• Curiobot LLM Tutoring Exploratory Learning — LLM tutoring for exploratory learning

Citation

Wang, C., Petrie, C., Stouras, M., Ettlin, N., George, A., Mejia-Domenzain, P., Swamy, V., Käser, T., & Svensson, O. (2026). Turning 500+ Students into Teachers: A Semester-Long Study of an AI Teachable Agent in an Undergraduate Algorithms Course. In Proceedings of the Thirteenth ACM Conference on Learning @ Scale (L@S '26).