ABOUT SESSION
We are entering a new era of building agentic applications—ones that not only understand context but plan, take action and adapt. With the advent of the Model Context Protocol (MCP) and a wave of orchestration frameworks, developers can now construct complex, long-running agentic systems with far greater modularity than ever before. But functionality is not enough — building "reliable" AI applications is the true challenge.
In this talk, I’ll explore what it means to build AI agents you can trust to act autonomously, while detecting, root-causing and acting quickly on errors. From defining “five nines” of agentic reliability to designing systems that can reason, revise, and recover gracefully, we’ll walk through the architectural principles that elevate agents from impressive prototypes to being dependable in production. No matter what agentic app you are building, the core question remains: can your AI agents be trusted to act correctly -- every single time?
Atin Sanyal
Сo-Founder and Chief Technology Officer, Galileo
