Peer Review Report For: Teacher-Mediated, Not Learner-Facing: A Bayesian Quasi-Experimental Study of Agentic AI Support in Grade 10 Functions and Graphs [version 1; peer review: 2 approved with reservations]

Abstract

Background Artificial intelligence (AI) is increasingly being integrated into school mathematics, yet most research has focused on learner-facing systems rather than AI that supports teachers during instruction. This study examined whether teacher-mediated agentic AI could improve Grade 10 learners’ achievement in functions and graphs while preserving teachers’ central instructional role. Methods A quasi-experimental study was conducted with 115 Grade 10 learners from two intact classes following the same eight-week instructional programme. One class received teacher-mediated AI-supported instruction, while the comparison class received conventional instruction. Bayesian baseline-adjusted models estimated differences in post-test achievement, mastery-threshold probabilities, and variation across baseline achievement levels. Learners’ mathematical interest and confidence were analysed as secondary outcomes. Results The Bayesian model estimated a baseline-adjusted advantage of 29.12 points for the AI-supported condition (95% credible interval: 27.31–30.92). Posterior estimates indicated substantially higher probabilities of reaching achievement thresholds of 70, 80, 90, and 95 under teacher-mediated AI-supported instruction than under conventional instruction. The estimated achievement contrast remained positive across the observed baseline achievement distribution. Learners in the AI-supported condition also demonstrated higher post-intervention interest and confidence than those receiving conventional instruction. Because the study employed two intact classes taught by different teachers, the findings should be interpreted cautiously as evidence from a quasi-experimental design rather than definitive causal effects. Conclusions Teacher-mediated agentic AI represents a distinct instructional configuration in which AI extends teachers’ pedagogical capacity without replacing professional judgement. Although the findings suggest considerable potential for supporting mathematics teaching, larger multi-teacher studies are needed to determine the robustness and generalisability of these effects.