This thesis investigates the feasibility of Large Language Model (LLM) agents as cross-functional orchestrators in Supply Chain Management (SCM), addressing the documented Reality Gap between sandbox experimentation and production-ready deployment. Building upon a structured literature review across eight supply chain domains, the study identifies critical limitations in semantic interoperability, governance, cross-agent coordination, and deployment robustness. To bridge these gaps, the thesis proposes a Hub-and-Spoke agentic reference architecture that integrates deterministic orchestration, domain-specialized agents, blockchain-based audit mechanisms, and a Human-in-the-Loop governance model. The methodological contribution consists of (i) a structured architectural framework for cross-functional LLM orchestration, (ii) a multi-dimensional evaluation model spanning operational, technical, and organizational metrics, and (iii) a conceptual deployment pathway outlining the conditions required for industrial readiness.