AI systems are rapidly evolving from passive models into autonomous agents capable of planning and executing workflows across cloud services and APIs. While advances in large language models enable powerful agent capabilities, deploying these systems in real production environments introduces new engineering challenges around reliability, safety, observability, and governance.

This talk explores agentic infrastructure the distributed systems required to safely run AI agents within cloud environments. We will examine how agent runtimes interact with tools, APIs, and microservices, and why traditional cloud reliability patterns are often insufficient for autonomous decision systems.

The session presents practical design patterns for building production-grade agent platforms, including action validation pipelines, guardrails, evaluation loops, and human-in-the-loop oversight. We will also discuss strategies for preventing cascading failures, controlling cost and behavior, and making agent decisions observable and auditable.

Attendees will leave with a systems-level blueprint for building reliable, safe, and observable AI agents in modern cloud infrastructure.