I wonder how the classical data science workflow of running test data experiments before got completely forgotten in the world of LLMs.