Foundations of Agentic AI for Retail - Second Edition: Operating Models, Decision Frameworks, and Retail‑Grade Architectures

Foundations of Agentic AI for Retail - Second Edition (Full-Color) Retail is where AI systems break or prove themselves: thin margins, live inventory, real customers, and continuous change. This second edition delivers the operating models, decision frameworks, and production practices needed to build agents that work—using retail as the rigorous, real-world test bed. What makes this edition essential? - Full-Color Visuals: 75+ diagrams, flowcharts, and architecture blueprints rendered in vivid color, making complex agent loops, data flows, and decision architectures immediately clear. - 50+ Retail Applications: Detailed coverage of demand forecasting, dynamic pricing, conversational merchandising, autonomous store operations, supply chain optimization, and customer experience—illustrating agent capabilities across the retail landscape. - 30+ Verified Code Examples: Hands-on Python listings covering BDI agents, OODA loops, MDPs, reinforcement learning, LLM-powered ReAct chains, MCP tool integration, A2A multi-agent orchestration, and AG-UI approval flows—all traced to a tested companion repository. - Retail-Grounded, Broadly Applicable: Master core decision frameworks (MDPs, Bayesian inference, causal reasoning) and reference architectures proven in retail, then adapt them to any high-constraint domain. - Foundations + Cutting-Edge Practice: Bridge classical AI (optimization, planning, RL) with modern breakthroughs: LLMs, OpenAI Agents SDK, structured outputs, retrieval-augmented generation (RAG), and interoperability protocols (MCP, A2A, A2UI). What you will learn: 1. Apply the Retail Agent Operating Model (RAOM) and three-plane architecture to design agents that sense, decide, act, and learn within policy constraints. 2. Use decision frameworks; Bayesian inference, causal reasoning, MDPs, and reinforcement learning, to solve real inventory, pricing, and customer experience problems. 3. Build multi-agent systems with Agents SDK, ADK, MCP, A2A, and A2UI protocols for tool integration, agent collaboration, and human-in-the-loop approvals. 4. Operate agents in production with SLOs, observability, CI/CD, and incident response playbooks. 5. Govern autonomy with evaluation harnesses, red-team playbooks, policy-as-code, audit trails, and escalation paths. New in the Second Edition: - Retail Agent Operating Model (RAOM) as the organizing spine - Three-plane architecture: capability, control, and integration planes - Contract-first schemas with verified, runnable companion code - Evaluation pyramid and red-team playbooks for safety testing - Agent learning ladder: prompt tuning → SFT/PEFT → DPO/GRPO → RLHF → RFT - Context engineering as a first-class design discipline - Interoperability snapshot: MCP, A2A, AG-UI protocols - Teaching resources, case packets, and notation guide Who this book is for: - Retail Leaders & Product Managers: Gain a clear roadmap for AI-driven transformation with governance and risk controls built in. - Software Architects & ML Engineers: Acquire hands-on guidance to design, build, and operate production-grade agentic platforms. - Researchers & Graduate Students: Use this as a rigorous yet practical reference bridging theory and deployment. - AI Practitioners: Explore the convergence of LLMs, computer vision, causal inference, and sensor networks in high-stakes retail environments. About the Author Dr. Fatih Nayebi is an Assistant Professor at McGill University and Founder & CEO of OODARIS AI. He has led enterprise-scale AI systems in retail and teaches graduate-level courses on agentic systems, decision intelligence, and AI operations.