Prompting an LLM is easy. Building a product that relies on one is something else entirely. In this talk, we’ll go beyond prompt hacking and look at what it takes to integrate LLMs into production systems with real users, real latency constraints, and real consequences. We’ll explore patterns like model orchestration, prompt versioning, skill/tool design, and how to connect your LLM to internal APIs, code search, and business logic.

You’ll also see practical strategies for observability, testing, and debugging — all battle-tested in real-world infrastructure. Whether you’re building a platform agent or just trying to make your product smarter, this talk will give you the mental model and technical tools to do it right.