一句话：提出生成式AI多智能体系统DevCoach，模拟开发团队角色支持SDLC学习，实验显示学习增益与任务完成率显著提升。
软件开发生命周期（SDLC）对软件质量至关重要，但学习 SDLC 概念面临协作、实时交互以及团队多技能组合等独特挑战。本文提出 DevCoach，一个由生成式 AI 驱动的多智能体系统，其中各智能体模拟软件开发团队中的不同角色，学生可与之交互并完成 SDLC 各阶段的任务。作者通过一项 20 名参与者的用户研究，评估系统对学习效果的提升、对 SDLC 交付物的影响，以及对有效学习环境所需的探究社区（CoI）要素的支持程度。结果表明，使用 DevCoach 的学生获得了显著更高的学习增益，并在所有 SDLC 阶段都有更高的任务完成率；系统还支持 CoI 要素，尤其是感知社会临场感。参与者称赞其即时且上下文感知的反馈、交互式学习环境以及多智能体团队提供的多样化专业知识。这些发现表明生成式 AI 有望通过更具协作性与实践性的学习体验，使软件工程教育更有效、更投入、更具交互性。
The software development life cycle (SDLC) is vital for ensuring the quality of software systems. However, learning SDLC concepts presents unique challenges, such as the need for effective collaboration, real-time interaction, and access to diverse skill sets represented in software development teams. To address these problems, we present DevCoach, a generative AI powered multi-agent system designed to support students learning the SDLC. DevCoach allows students to interact with generative AI agents simulating the different roles in the software development team, engaging in tasks across different phases of SDLC. Through a user study (n = 20), we evaluate the system's effectiveness in enhancing learning, impact on SDLC deliverables, and support for Community of Inquiry (CoI) elements necessary for effective and supportive learning environments. Our results reveal that students using DevCoach achieved significantly higher learning gains and improved task completion rates across all SDLC phases. The system also supports CoI elements, particularly perceived social presence. Participants also lauded the immediate context-aware feedback, interactive learning environment, and diverse expertise provided by the roles within the multi-agent team. These findings demonstrate the potential of generative AI to enhance software engineering education by making it more effective, engaging, and interactive, providing students with collaborative and practical learning experiences.
@inproceedings{paperbot9046,
title = {DevCoach: Supporting Students Learning the Software Development Life Cycle with a Generative AI powered Multi-Agent System},
author = {Tianjia Wang and Matthew Trimble and Chris Brown},
booktitle = {FSE},
year = {2025},
doi = {10.1145/3696630.3727255}
}