Contrary to viewing AI primarily as learning support, this research treats it as an agency for mediating discourse, thereby redefining participants’ involvement, feedback mechanisms, sources of authority, learner agency, dependency, authorship, and legitimacy of acquired knowledge. A mixed-methods approach was employed in this study, with two-hundred and fifty Saudi EFL undergraduates majoring in English at four Saudi universities answering a questionnaire rooted in the frameworks of SFL and CDA and interviews with a purposive sample of twenty participants. Descriptive statistics, reliability, correlation, and one-way ANOVA were used to analyze quantitative data while qualitative data was analyzed thematically using notions derived from SFL and CDA frameworks. Results indicated that students perceived AI-mediated interaction as instrumental in shifting classroom discourse from traditional teacher-student structure to a triadic teacher-student-AI structure. Moreover, AI was seen as a private tutor, feedback provider, corrector, and linguistic evaluator, whereas teachers were still considered necessary for contextual validity, guidance on ethics, and pedagogical judgment of learners’ performance. Gains were reported by the students’ during individual usage of AI as a learning agency, though reservations were expressed about overdependence on AI, decreased interaction with peers/teachers, standardization of language use, and doubtful authorship. The study concluded that the role of AI in Saudi EFL classrooms went beyond being merely supportive in education and shaped the restructuring of discourse relations, power structures, and language learning practices.