一句话：提出 AI 驱动软件开发课程及多维度评估模型，以过程、代码与反思报告衡量学生在 LLM 时代的学习。
大语言模型正在改变软件开发，传统以代码功能为标准的评分方式因 AI 生成代码而失效，学生也需系统了解 AI 工具在软件生命周期中的潜力与风险。作者设计“AI 驱动软件开发”课程：先以脚手架式教学覆盖开发阶段与相应 AI 工具，再让学生独立借助 AI 技术构建复杂软件系统；评分综合最终代码、开发过程与反思报告，旨在评估学生学习而非 AI 能力。在 68 名学生的首次实施中，学生成功构建了非平凡系统，评分方案还揭示了对何时及如何有效使用 AI 工具的深入反思，为未来课程和 AI 工具开发提供洞见。
The proliferation of Large Language Models (llms) is transforming the way software is developed, and software engineering education needs to adapt to this new reality. Traditional grading methods that assess code functionality are undermined by ai-generated code. Meanwhile, students are already using these tools and require formal training on their potential and dangers throughout the software lifecycle. We address this with a new course on AI-Driven Software Development. After an initial scaffolded phase covering software development phases and corresponding ai tools, students independently build a complex software system using ai techniques. To evaluate student learning rather than ai capabilities, our grading is based on the final code, the development process, and a reflective report. In our first implementation of this course with 68 students they not only successfully built non-trivial software systems with ai support—our grading scheme also reveals deeper reflections on when and how to effectively use ai tools, which also provide valuable insights informing future courses on ai-driven software development as well as ai tool development in general.
@inproceedings{paperbot11994,
title = {AI-Driven Software Development: A New Course Concept and Assessment Model for the Era of Large Language Models},
author = {Benedikt Fein and Gordon Fraser and Steffen Herbold},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3786580.3786962}
}