How do you achieve reliable results from stochastic AI models? This talk presents practical techniques for building quality pipelines around generative AI that improve output quality before human review begins, making more effective use of limited human time and attention.