AltDev: Achieving Real-Time Alignment in Multi-Agent Software Development
Keywords: Multi-Agent; Large Language Model; Software Development;
Abstract: Large Language Models (LLMs) have shown remarkable capability in code generation tasks. However, they still struggle with complex software development tasks where agents of different roles need to work collaboratively. Existing works have proposed some LLM-based multi-agent software development frameworks following linear models such as the Waterfall model. However, linear models suffer from erroneous outputs of LLMs due to the lack of a self-correction mechanism. Inspired by human teams where people can freely start meetings for reaching agreement, we propose a novel and flexible multi-agent framework AltDev, which enables agents to correct their deliverables and align with other agents in a real-time manner. AltDev integrates a compulsory alignment checking and a conditional multi-agent discussion at the end of each development phase, in order to identify and reduce errors at early stages in the software development lifecycle. Our experiments on various software development tasks show that AltDev significantly improves the quality of generated software code in terms of executability, structural and functional completeness.
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8651
Loading