This presentation explores a practical, classroom-tested approach to working with AI rather than against it in Oral English courses. Acknowledging that students already use AI tools, we argue that technological change reshapes learning demands rather than eliminating learning itself. A common classroom challenge is that students often focus on producing a “perfect” written script for presentations rather than developing spoken fluency. This emphasis encourages memorization, which can increase anxiety and reduce students’ confidence when speaking spontaneously. To address this issue, we redesigned a speaking task that prioritizes human cognition before AI involvement. Students first research authentic data, develop their own ideas and sentences, and organize speaking notes independently. Only after this stage is AI introduced through clearly bounded tools rather than a single solution. NotebookLM supports source-based research within transparent constraints, while ChatGPT provides prompt-driven language scaffolding focused on clarity, structure, and delivery rather than content generation. Together, these tools form an ethical workflow that reduces anxiety and improves speaking performance. The presentation concludes by discussing implications for assessment design, AI literacy, and teacher mediation in CALL, emphasizing that effective pedagogy lies not in banning AI, but in teaching students when, why, and how to use it responsibly and ethically.