Multi-Agent Software Development through Cross-Team Collaboration

Multi-Agent Software Development through Cross-Team Collaboration

13 Jun 2024 | Zhuoyun Du, Chen Qian, Wei Liu, Zihao Xie, Yifei Wang, Yufan Dang, Weize Chen, Cheng Yang
The paper introduces Cross-Team Collaboration (CTC), a scalable multi-team framework designed to enhance software development through cross-team collaboration. CTC enables multiple teams to jointly propose various decisions and communicate insights, leading to superior content generation. The framework addresses the limitations of single-agent teams, which can only execute tasks sequentially and lose opportunities to explore multiple decision paths. Through experiments on the SRDD dataset for software generation and the ROCStories dataset for story generation, CTC demonstrates significant improvements in quality compared to state-of-the-art baselines. The framework's effectiveness is attributed to its ability to facilitate diverse content exchange, effective pruning of low-quality content, and hierarchical partitioning of teams. The paper also discusses limitations, such as the potential loss of valuable insights through pruning and the need for precise software requirements. Overall, CTC shows promise in enhancing the quality of software and story generation tasks, suggesting a shift towards multi-team collaboration in LLM agent development.The paper introduces Cross-Team Collaboration (CTC), a scalable multi-team framework designed to enhance software development through cross-team collaboration. CTC enables multiple teams to jointly propose various decisions and communicate insights, leading to superior content generation. The framework addresses the limitations of single-agent teams, which can only execute tasks sequentially and lose opportunities to explore multiple decision paths. Through experiments on the SRDD dataset for software generation and the ROCStories dataset for story generation, CTC demonstrates significant improvements in quality compared to state-of-the-art baselines. The framework's effectiveness is attributed to its ability to facilitate diverse content exchange, effective pruning of low-quality content, and hierarchical partitioning of teams. The paper also discusses limitations, such as the potential loss of valuable insights through pruning and the need for precise software requirements. Overall, CTC shows promise in enhancing the quality of software and story generation tasks, suggesting a shift towards multi-team collaboration in LLM agent development.
Reach us at info@study.space
[slides] Multi-Agent Software Development through Cross-Team Collaboration | StudySpace