Agentic systems powered by LLMs don’t behave like traditional software. They’re non-deterministic, make decisions on the fly, and sometimes surprise you. So what does reliability even mean in a world like that?

In this session, we’ll take a fresh look at familiar SRE practices like SLOs and error budgets, and see how they hold up when your system doesn’t always give the same answer twice. We’ll also explore how chaos engineering can help you uncover failure modes in agent workflows before they show up in production.

You’ll walk away with practical ideas to bring more control and confidence to AI-driven systems, without killing the speed and flexibility that make them powerful.