AI agents are only as reliable as the models they depend on, but what happens when capacity is maxed out, latency spikes, or a model becomes unavailable? In this keynote, we’ll explore SRE-grade approaches to agent maintenance and multi-LLM fallback strategies. You’ll learn how to design AI systems that gracefully switch between models, maintain reliability under pressure, and deliver consistent performance even when your primary model is unavailable. Think of it as applying tried-and-true SRE principles of failover, observability, and incident readiness now in the world of AI.
Rohan Gupta is Principal Product Manager at Harness, where he leads the AI transformation of DevOps through innovations like the Harness AI DevOps Agent, Unified Agent, and MCP Server. Over six years at Harness, he has driven the evolution of Continuous Delivery and NextGen platforms, scaling adoption, launching core integrations, and redefining how engineering teams build and manage software delivery pipelines.
