DSPy is Stanford NLP's programming model for algorithmically optimizing LLM prompts and weights, treating AI pipelines as programs that can be compiled and optimized rather than manually tuned strings of text. Instead of hand-crafting prompts, DSPy developers write declarative programs specifying what their AI system should do, and DSPy's optimizer automatically generates and refines the best prompts for any LLM. This approach dramatically reduces prompt engineering effort, makes pipelines more maintainable, and enables systematic improvement through automated optimization loops.