This study aims to explore how EFL learners experience and manage speaking anxiety while engaging in AI-mediated speaking practice using MoodieAI. A narrative inquiry design was employed, guided by Horwitz, Horwitz, and Cope's (1986) Foreign Language Classroom Anxiety (FLCA) framework and interpreted through Clandinin's (2013) three-dimensional narrative inquiry space. The participants were three undergraduate EFL students who had used MoodieAI for at least three months. Data were collected through semi-structured interviews, supported by digital artifacts for triangulation, and analyzed using narrative thematic analysis. The findings reveal that speaking anxiety in this AI-mediated context remained multidimensional, encompassing communication apprehension, fear of negative evaluation, and test anxiety, with discomfort most intense during participants' earliest encounters and partly rooted in the physical and social recording environment rather than the AI itself. The findings also show that the same features of MoodieAI that initially triggered anxiety gradually supported its management through self-regulation strategies and growing confidence across repeated practice. These findings confirm that the FLCA framework remains relevant in AI-mediated contexts while revealing nuances that extend it beyond traditional classroom settings.