# Communicative Agents for Software Development

**Authors:** Chen Qian, Xin Cong, Wei Liu, Cheng Yang, Weize Chen, Yusheng Su, Yufan Dang, Jiahao Li, Juyuan Xu, Dahai Li, Zhiyuan Liu, Maosong Sun  
**arXiv:** [2307.07924](https://arxiv.org/abs/2307.07924)  
**Date:** 2023-07-16  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2307.07924v3/

## Key points

- Recent advancements in deep learning are revolutionizing software development practices
- Large language models (LLMs) are used throughout the entire software development process
- ChatDev is a virtual chat-powered software development company that follows the waterfall model
- The development process is divided into four stages: designing, coding, testing, and documenting
- Each stage involves a team of agents who engage in collaborative dialogue facilitated by the chat chain
- ChatDev enables efficient resolution of specific subtasks through context-aware communication
- ChatDev completes the entire software development process in under seven minutes at a cost of less than one dollar
- ChatDev identifies and addresses potential vulnerabilities and rectifies potential hallucinations
- Integrating LLMs into software development improves decision-making processes, efficiency, and cost-effectiveness.

## Summary

In the field of software engineering, decision-making processes often require intuition and consultation. Recent advancements in deep learning have begun to revolutionize software development practices by incorporating elaborate designs at various stages. This paper introduces a novel paradigm that utilizes large language models (LLMs) throughout the entire software development process, streamlining and unifying key processes through natural language communication. This eliminates the need for specialized models at each phase. At the center of this paradigm is ChatDev, a virtual chat-powered software development company that follows the established waterfall model. The development process is divided into four distinct chronological stages: designing, coding, testing, and documenting. Each stage involves a team of agents including programmers, code reviewers, and test engineers who engage in collaborative dialogue facilitated by the chat chain. The chat chain breaks down each stage into atomic subtasks, enabling dual roles for proposing and validating solutions through context-aware communication. This leads to efficient resolution of specific subtasks. An instrumental analysis of ChatDev demonstrates its remarkable efficacy in software generation as it enables the completion of the entire software development process in under seven minutes at a cost of less than one dollar. Additionally, ChatDev not only identifies and addresses potential vulnerabilities but also rectifies potential hallucinations while maintaining efficiency and cost-effectiveness. The potential of ChatDev opens up new possibilities for integrating LLMs into the realm of software development with improved decision-making processes that are more efficient and cost-effective than ever before.