AI-Powered Multi-Agent System Enhances Collaborative Decision Making in Distributed Environments

Abstract

In applications like Industry 4.0 cyberphysical systems, self-organizing networks and edge-efficient IoT deployments the data that are generated are dynamic, voluminous, and require real-time, adaptive and collaborative decision making which is not possible with centralized or traditional rule-based approaches. Existing solutions, though they have made significant strides in multi-agent systems and orchestration through large language models, lack the ability to integrate high-level reasoning with scalable coordination, have a lack of governance control, and are not robust enough to deal with real world variability, leading to problems of coordination failures, latency bottlenecks, and accountability gaps. This paper introduces a novel hybrid multi-agent framework that combines LLM orchestration with specialized agents with specific roles for perception, analytics, optimization, and decision at multiple layers and enables multi-round collaborative reasoning, dynamic feature-role alignment and built-in coordination transparency with logging, monitoring, and human-in-the-loop intervention hooks. When tested on datasets of manufacturing prescriptive maintenance tasks and network traffic, the proposed system achieves better performance with higher accuracy 0.78, precision 0.80, recall 0.75 and F1-Score 0.77 as compared to strong baselines such as LaMA-SON, hybrid agentic methods with 8-15% improvement across various metrics and exhibits lower latency and higher robustness to noise and distribution shift. These findings provide a technically and sociotechnical groundwork for the trustworthy collaboration of intelligences in distributed smart environments, which can be easily expanded.