Engineering Patterns for Building AI-native applications

AI agents become much harder to engineer once they move beyond a single model call. This session explores practical patterns for building production-ready agentic applications, including orchestration strategies, managing what enters the context window, structuring model and tool calls, and recovering from partial failures.

We’ll look at how to design workflows that can retry safely, resume after interruptions, preserve the right state, and avoid unnecessary context and model usage. The focus is on concrete engineering patterns for building AI-native applications that are reliable, efficient, and maintainable in production.