Agyn: A Multi-Agent System for Team-Based Autonomous Software Engineering

TL;DR

The paper tackles autonomous software engineering by reframing it as a team-based organizational task and introducing a production-oriented, multi-agent system built on the agyn platform. By allocating distinct roles (manager, researcher, engineer, reviewer), providing isolated execution environments, and enforcing a GitHub-native, automation-focused workflow, the approach demonstrates competitive performance on SWE-bench 500 without benchmark-specific tuning. Key contributions include the agyn-based framework, role-specific model configurations, automated inline reviews and PR management tooling, and a production deployment perspective highlighted by the 72.4% automated resolution rate. The work underscores the importance of organizational design, role separation, and coordinated workflows in advancing autonomous software engineering beyond model-centric improvements.

Abstract

Large language models have demonstrated strong capabilities in individual software engineering tasks, yet most autonomous systems still treat issue resolution as a monolithic or pipeline-based process. In contrast, real-world software development is organized as a collaborative activity carried out by teams following shared methodologies, with clear role separation, communication, and review. In this work, we present a fully automated multi-agent system that explicitly models software engineering as an organizational process, replicating the structure of an engineering team. Built on top of agyn, an open-source platform for configuring agent teams, our system assigns specialized agents to roles such as coordination, research, implementation, and review, provides them with isolated sandboxes for experimentation, and enables structured communication. The system follows a defined development methodology for working on issues, including analysis, task specification, pull request creation, and iterative review, and operates without any human intervention. Importantly, the system was designed for real production use and was not tuned for SWE-bench. When evaluated post hoc on SWE-bench 500, it resolves 72.4% of tasks, outperforming single-agent baselines using comparable language models. Our results suggest that replicating team structure, methodology, and communication is a powerful paradigm for autonomous software engineering, and that future progress may depend as much on organizational design and agent infrastructure as on model improvements.