Cyber Times International Journal of Technology & Management

The emergence of large language model (LLM)-based multi-agent systems has initiated a fundamental reconfiguration of enterprise automation paradigms. Where single-model pipelines once governed  discrete task execution, coordinated networks of autonomous agents now negotiate complex, multi-step workflows across heterogeneous organisational environments. Yet, despite accelerating deployment across sectors including finance, healthcare, and logistics, the comparative capabilities, governance readiness, and scalability of leading multi-agent LLM frameworks remain poorly understood in rigorous scholarly terms. This study conducts a systematic comparative evaluation of four prominent frameworks LangGraph, AutoGen, CrewAI, and OpenAI Swarm — assessing their architectures, orchestration mechanisms, tool-use capabilities, and alignment with enterprise governance standards. Employing a Design Science Research (DSR) methodology enriched by structured scenario-based analysis, the study proposes a tiered evaluation rubric grounded in real-world workflow complexity. Findings reveal significant architectural divergence between frameworks optimised for deterministic orchestration and those designed for emergent autonomy, with meaningful implications for enterprise risk posture, scalability, and regulatory compliance. The study contributes a reusable enterprise readiness framework for multi-agent LLM deployment, contextualised against OECD AI Principles, NIST AI Risk Management Framework (AI RMF), and emerging EU AI Act provisions. Results offer actionable guidance for technology leaders, policymakers, and researchers navigating agentic AI adoption at institutional scale.