Maxime Rivest and Isaac Miller present DSPy's approach to reusable AI programs: keep inputs, outputs and task requirements stable while changing models, prompts and implementation techniques. They distinguish natural-language specifications, deterministic constraints and examples that define quality, using tax extraction and other repeatable workflows to illustrate the boundary.
Maxime Rivest and Isaac Miller discuss enterprise examples and techniques including recursive language models, prompt optimisation and DSPy Flex. They describe qualitative learning as an active research direction for turning production feedback into evolving evaluations, while stressing that stronger models still need task-specific context and measurable outcomes.
Watch on YouTube
