Have you written an LLM skill? It starts simple: define its goal, give it a tool, and it works. But what happens when you drop that isolated skill into a live system with hundreds of capabilities? At Wix, handling 300K monthly conversations, the challenge isn't just writing a skill—it's finding the right one in a sea of options. In this session, we’ll explore the lifecycle of production LLM skills. First, we’ll define what should actually be a skill and where they commonly crash in the wild (e.g., missing exit points, outdated context). Then, we’ll dive into the routing dilemma: When you have hundreds of actions, how do you trigger the exact right one? Do you retrieve them via RAG? Shove them all into the context window? Or filter them using deterministic code? Join us to compare these architectural approaches and learn how to scale autonomous agents effectively.
Senior Data Scientist @ Wix