一句话：通过持续交互式考核改革全栈开发课程，使学生在 AI 辅助下仍能展示对交付物的真正掌握。
生成式 AI 降低了全栈开发成本，使“系统能跑即掌握知识”的传统假设失效。本文针对软件工程课程中稀疏检查点与小组评估导致的“功能性完成”失败模式，开展干预研究。作者不禁止 AI，而是引入持续交互式评估框架：通过频繁检查点、现场解释和过程证据，让学生无论是否使用 AI 都能证明对交付物的所有权。评估采用含 44 个核心问题的 SE 教育问卷与课程单元调查。结果显示学生 AI 使用显著转向细粒度与可验证：75.7% 会将任务分解为更小提示，77.6% 知道何时停止提示并转入手动编码，75.7% 能解释 AI 辅助方案为何有效。研究表明 AI 时代软件工程教育需让学习过程可见、可解释、可评估。
Generative AI has reshaped software engineering education by lowering full-stack development cost for production, weakening the traditional assumption that a working system implies working knowledge. This paper presents an intervention study that targets a recurring failure mode in traditional software engineering courses, where sparse checkpoints and group-only assessments allow students to achieve "functional completion" via AI with minimal reasoning. Rather than prohibiting AI, we introduce a continuous interactive assessment framework through frequent checkpoints with live explanations and process evidence, so that students can demonstrate deliverable ownership with or without AI assistance. We evaluate the redesigned offerings using (i) a SE Education questionnaire with 44 core questions and (ii) unit of study surveys. Results indicate a significant shift toward fine-grained and verifiable AI usage. 75.7% of respondents report decomposing tasks into smaller prompts, 77.6% report knowing when to stop prompting and switch to manual coding, and 75.7% report being able to explain why an AI-assisted solution works. Our findings suggest that re-aligning software engineering education in the AI era requires making learning processes visible, explainable, and assessable.
@inproceedings{JinWBHC26,
title = {Deliverables Are Not Understanding in the AI Era: Reforming Full-Stack Development Education with Continuous Interactive Learning},
author = {Haolin Jin and Jiawen Wen and Zhaoge Bi and Linghan Huang and Huaming Chen},
booktitle = {FSE},
year = {2026},
doi = {10.1145/3803437.3805794}
}