The hype around AI coding agents is very real, and autonomous coding systems are improving fast. In this talk, we'll explore the "Ralph Wiggum" method, where AI agents run in persistent iteration cycles until tasks are actually complete. You will learn about backpressure mechanisms to let the LLM self-correct and retry without human intervention. We will discuss how to structure the prompts, practical patterns for turning test failures into actionable AI feedback, and honest insights about when this approach works best versus where it falls flat. This is about making AI agents actually useful for real work by applying agentic engineering principles, not vibe coding. If you're curious about the future of AI-assisted development and want to see what's possible when we design systems that expect and handle failure, this talk is for you.
