从任务委托到协作迁移：生成式人工智能支持新托福听力备课的实践反思——以“Listen to an Announcement”课程开发为例

From Task Delegation to Collaborative Transfer: Practical Reflections on Agentic AI-Supported TOEFL iBT Listening Lesson Preparation—A Case Study of “Listen to an Announcement” Course Development

摘要: 生成式人工智能(Agentic AI)正在进入教师备课、资源开发与课程设计等日常专业活动，但已有讨论多集中于内容生成效率、提示词技巧或使用风险，对教师与AI在持续互动中如何共同推进问题、修正知识的过程关注仍显不足。文章以笔者使用Codex准备新托福听力“Listen to an Announcement”课程的一次完整实践为案例，基于53篇不重复听力文本、105道题目、连续人机对话、AI生成的系列备课文档以及最终形成的379页课件，选取三个关键事件进行反思性分析。研究发现：第一，Agentic AI能够承担文件读取、任务拆解、语料统计和关联文档更新等长链条工作，使教师得以从逐篇处理材料转向观察整体规律；第二，AI产出并非只是问题的答案，也可能使原本隐而未显的现象变得可见，进而触发教师提出新的研究问题；第三，AI的专业误判往往具有表面合理性，并可能借助自动化能力扩散至统计、规律总结和例题选择等多个环节。文章据此提出“任务委托–结果检视–问题生成–再分析–专业审查–教学转化”的人机协作路径。

Abstract: Agentic AI is increasingly entering teachers’ everyday professional practices, including lesson preparation, resource development, and course design. Existing discussions, however, have often focused on content-generation efficiency, prompt engineering, or potential risks, while paying less attention to how teachers and AI systems jointly advance questions and revise knowledge through sustained interaction. This article presents a reflective case study of the author’s use of Codex to prepare a TOEFL iBT Listening course on “Listen to an Announcement”. The analysis draws on 53 unique listening texts, 105 questions, continuous human-AI dialogues, a series of AI-generated lesson-preparation documents, and a 379-page courseware product developed from these materials. Three key events are examined during this process. The study finds that: First, Agentic AI can support long-chain work such as file reading, task decomposition, corpus-based statistics, and cross-document updating, thereby allowing teachers to move from item-by-item processing to the observation of broader patterns. Second, AI outputs may function not only as answers but also as prompts that make previously unnoticed problems visible. Third, AI errors may appear superficially reasonable and can spread quickly across statistics, pattern summaries, and teaching examples through automated updating. The article therefore proposes a human-AI collaboration pathway of “task delegation, result inspection, question generation, re-analysis, professional review, teaching transformation, demonstration feedback, and collaborative transfer”.