LangChain has published a guide to regression testing with LangSmith for LLM applications. The workflow helps teams compare experiments, track performance, and identify behavior changes as prompts, models, and retrieval systems evolve.
Regression testing is especially important for LLM apps because improvements in one area can quietly break another. A new prompt or model may look better in demos while degrading edge cases or product-specific expectations.
The guide reflects a maturing development process for AI products: teams need repeatable tests and history, not just manual spot checks.