Over the last two years, GenAI took the world by storm, and disrupted the way people interact with software and create content. Quoting Andrej Karpathy: “The hottest new programming language is English” Klaviyo adopted GenAI early on, and today 12+ of our features leverage it in production. We've been through the highs and lows to augment our features with reactive and proactive experiences that assist our customers in object generation, content creation, summarization or contextual help. In this talk, we will discuss our approaches, learnings and stack, and how it has evolved over time for each of these GenAI app archetypes. We will cover the importance of evals-driven-development, structured outputs, how they can add reliability and security, how we’ve mitigated certain risks of GenAI applications, how we tackled internationalization to support our market expansion, the evolution of our stack and how we continue to optimize for flexibility to adapt to a fast-changing environment. Additionally, we will talk about our approaches to RAG and how we’ve enhanced our RAG pipeline by adopting different embedding strategies as well as combining hybrid search patterns effectively. Last, we will talk about the importance of the true differentiator, the data, and why we’re excited about the future.