A Scalable Multi-Agent Architecture For Agentic AI-Based Autonomous Task Coordination and Resource Optimization

Abstract

The study establishes a scalable multi-agent architecture for agentic artificial intelligence (Agentic AI) systems that can support autonomous task coordination, distributed execution, and efficient computational resource utilization under changing workload conditions. The objective is to integrate agent-level reasoning, task planning, autonomous task allocation, execution monitoring, resource management, and load balancing within a unified modular framework. The study also examines task-allocation behaviour under different problem scales using established allocation approaches, including Hungarian, Auction, Consensus-Based Bundle Algorithm (CBBA), and Auction 2-opt. The evaluated benchmark scenarios indicate that task scale influences computational requirements and communication overhead. At small scale, the Hungarian method achieved the lowest total cost, while Auction 2-opt required the highest computational effort and communication. At medium scale, the Auction method produced the lowest total cost, whereas Auction 2-opt again showed greater computational requirements and communication overhead. At large scale, differences in total cost remained relatively small, but computational requirements increased considerably with problem size. These findings demonstrate that allocation performance varies according to workload scale and that resource-aware coordination is important for scalable multi-agent execution. The proposed architecture provides a structured foundation for integrating autonomous reasoning, multi-agent coordination, task allocation, and resource optimization. The benchmark evidence identifies important considerations for future implementation and supports further validation through original simulations and real-world experiments involving dynamic workloads, heterogeneous resources, failures, and distributed edge-cloud environments.