Large language models (LLMs) are increasingly being used by companies to enhance product experiences and internal operations, marking a shift in the computing landscape. Evaluating LLMs presents unique challenges due to their complexity and the subjective nature of their outputs, differing from traditional software and machine learning models. The evaluation process involves various components such as prompt templates, data sources, and memory, all of which require careful configuration. Testing LLMs often focuses on integration and end-to-end tests instead of unit tests, due to factors like randomness, subjectivity, and scope. Observability and monitoring are evolving to suit the needs of LLM applications, which benefit from rapid iteration and input from diverse teams. Evaluation strategies include leveraging human, model, and heuristic judgments, with model judgments gaining prominence due to their scalability. High-quality datasets are crucial, and can be sourced from real user interactions or synthesized using LLMs. This dynamic field continues to advance, with future developments expected in AI-based evaluators, multi-modal applications, and complex agent-based workflows.