A System Approach To Architecting Multi-Agent Systems In Enterprises

Abstract

This thesis develops a systematic framework for the architecture, evaluation, and phased deployment of AI-powered multi-agent systems in enterprise environments. Beginning with a comprehensive review of agent capabilities, architectural patterns, coordination protocols, and governance models, the study identifies critical design criteria guiding effective MAS integration. Employing a structured systems engineering approach, combining architectural decision analysis, multi-attribute utility modeling, and Monte Carlo-based tradespace exploration, the research quantifies cost-performance trade-offs and evaluates MAS concepts against prioritized enterprise metrics. Prototype implementations targeting real-world workflow challenges validate key architectural choices and inform a detailed strategic and technical roadmap for MAS integration, grounded in an AI System Readiness Level framework. This thesis concludes with actionable recommendations and future research directions to accelerate the adoption and maturity of MAS in organizational practice.