MULTI-AGENT AI SYSTEMS FOR AUTONOMOUS SOFTWARE DEVELOPMENT

Autonomous artificial intelligence (AI) agents, Large Language Models (LLM), and the use of multi-agents are quickly changing the landscape of software engineering by making more software development methods more automated.With the increasing complexity of software systems, there is an increasing demand of intelligent systems that can automatically aid requirements analysis, software design, code generation, testing, debugging, documentation, deployment and maintenance.A promising trend has been to delegate these tasks to Multi-Agent AI Systems which allocate these duties to autonomous agents which communicate, plan tasks, examine the output of other autonomous agents and enhance their software artifacts through iterative improvement.This paper explores the role and capabilities of Multi-agent AI in autonomous software engineering, along with the benefits and challenges in the form of a Systematic Literature Review (SLR) and comparative analysis of recent studies on the autonomous software development using a LLM and multi-agent systems.The results suggest that multi-agent systems can allocate software engineering tasks among specialized agents, which eliminates the reliance on a single AI system and can serve the purpose of requirements analysis, planning, code generation, testing, debugging, code review and documentation (Rasheed et al., 2024; Qian et al., 2024).Examples of such frameworks as ChatDev, AutoGen Studio, AutoDev, CodePori, and Magentic-One show various methods of coordinating agents and performing independent tasks (Tufano et al., 2024;Dibia et al., 2024).Task separation, software quality assurance, and workflow automation can be enhanced with the help of collaborative agent architectures, but their efficiency will largely rely on communication standards, coordination, context management, and evaluation objectives (Talebirad and Nadiri, 2023; Manish, 2024).The common obstacles consist of hallucinated/untrusted code, unreliable inter-agent communication, physical security threats, computational expenses, little human control, and the inability to confidently assess autonomous-development procedures (Suri et al., 2023).In general, Multi-Agent AI has a great potential to transform AI-assisted software engineering to more autonomous development cycles, yet careful validation, proper governance, and meaningful human oversight is a necessary requirement to ensure effective and safe adoption.