Agentic Commerce at Scale: A Reference Architecture for Enterprise Retail Systems

Abstract

Autonomous shopping and operations through specialized agent networks represent the next evolution in retail, yet most implementations remain constrained to experimental deployments due to fundamental gaps in safety, governance, and enterprise integration. This article presents a comprehensive cloud-native reference architecture that enables production-grade agentic commerce across the retail technology stack. The architecture encompasses buyer-side agents for discovery, negotiation, and checkout operating at the edge and in cloud environments, alongside merchant-side agents for catalog intelligence, dynamic promotions, and service automation, all governed by policy-enforced guardrails. A novel orchestration layer provides model routing, tool access control, memory management, and multi-stage safety filtering, while an integrated observability plane tracks reliability, cost, and risk metrics aligned to service-level objectives. The framework addresses critical enterprise constraints, including PII protection, brand safety, regulatory compliance, and seamless integration with existing product information management, order management, inventory, and content systems. Three production case studies demonstrate practical deployment patterns for promotion negotiation, returns automation, and catalog question-answering, establishing validated pathways from experimental prototypes to enterprise-scale autonomous commerce platforms.