Mitigating Collaboration Degeneration in Multi-Agent Code Generation via a Controllable Competitive Collaboration Approach

Gu, Shanzhi; Geng, Mingyang; Dong, Yihong; Mao, Yunxin; Qu, Zhaoyang; Zou, Hao; Jin, Ruochun; Liu, Zhipeng; Xu, Chuanfu; Wang, Haotian

doi:10.24963/ijcai.2026/15

Mitigating Collaboration Degeneration in Multi-Agent Code Generation via a Controllable Competitive Collaboration Approach

Shanzhi Gu, Mingyang Geng, Yihong Dong, Yunxin Mao, Zhaoyang Qu, Hao Zou, Ruochun Jin, Zhipeng Liu, Chuanfu Xu, Haotian Wang

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 127-135.

https://doi.org/10.24963/ijcai.2026/15

Keywords

Agent-based and Multi-agent Systems: Multi-agent learning
Agent-based and Multi-agent Systems: Coordination and cooperation

Abstract

Empowered by large language models (LLMs), multi-agent systems (MAS) have shown significant potential in code generation by simulating collaborative workflows. However, we identify a collaboration degeneration phenomenon, where one agent dominates while others remain disengaged, occurring in 38.4% of non-routine HumanEval tasks. Surprisingly, introducing competition among LLM agents eliminates this degeneration and sparks innovation in generated code. Inspired by this catfish effect, we posit that competition helps prevent collaboration degeneration. We thus propose C3, a Controllable Competitive Collaboration framework featuring two novel mechanisms: Centralized Auction-Based Competition (CAB) for role allocation via bidding and

Decentralized Communication-Aware Competition (DCC) for solution refinement through opponent-aware adaptation.

Evaluated on 70 complex projects from extended SoftwareDev, C3 reduces collaboration degeneration from 32.4% to 12.8%, while enhancing code innovation and diversity. Human evaluations further confirm C3’s superiority in functionality, maintainability, and robustness over purely collaborative or competitive baselines.

Shanzhi Gu, Mingyang Geng, Yihong Dong, Yunxin Mao, Zhaoyang Qu, Hao Zou, Ruochun Jin, Zhipeng Liu, Chuanfu Xu, and Haotian Wang. 2026. Mitigating Collaboration Degeneration in Multi-Agent Code Generation via a Controllable Competitive Collaboration Approach. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 127–135. https://doi.org/10.24963/ijcai.2026/15

@inproceedings{ijcai2026p15,
  title     = {Mitigating Collaboration Degeneration in Multi-Agent Code Generation via a Controllable Competitive Collaboration Approach},
  author    = {Gu, Shanzhi and Geng, Mingyang and Dong, Yihong and Mao, Yunxin and Qu, Zhaoyang and Zou, Hao and Jin, Ruochun and Liu, Zhipeng and Xu, Chuanfu and Wang, Haotian},
  booktitle = {Proceedings of the Thirty-Fifth International Joint Conference on
               Artificial Intelligence, {IJCAI-26}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Diego Calvanese},
  pages     = {127--135},
  year      = {2026},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2026/15},
  url       = {https://doi.org/10.24963/ijcai.2026/15},
}

TY  - CPAPER
TI  - Mitigating Collaboration Degeneration in Multi-Agent Code Generation via a Controllable Competitive Collaboration Approach
AU  - Gu, Shanzhi
AU  - Geng, Mingyang
AU  - Dong, Yihong
AU  - Mao, Yunxin
AU  - Qu, Zhaoyang
AU  - Zou, Hao
AU  - Jin, Ruochun
AU  - Liu, Zhipeng
AU  - Xu, Chuanfu
AU  - Wang, Haotian
T2  - Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2026
SP  - 127
EP  - 135
DO  - 10.24963/ijcai.2026/15
UR  - https://doi.org/10.24963/ijcai.2026/15
ER  -