When Agentic AI Fails in Production: Designing Robust Agentic Systems on AWS

In 2025, Agentic AI has demonstrated its ability to address a wide range of business domains — from financial services to marketing and retail. As successful pilots and early deployments deliver tangible value, organizations are increasingly pushing agentic systems from experimentation into production environments.

While Agentic AI promises autonomy and advanced reasoning, it also introduces a new class of production risks that traditional software systems — and even earlier Generative AI applications — do not face. These risks often surface only after early success, when autonomous behavior, non-determinism, and system-level interactions begin to scale.

In this talk, we examine the most common failure modes observed in real-world agentic AI deployments and discuss how to address them through robustness-aware system design. We explore architectural patterns, guardrails, and observability layers that help teams move beyond fragile prototypes and build agentic systems that are reliable, governable, and production-ready on AWS.