An engineering team is building two applications using the same language model. Application A is a real-time chatbot where users expect fast, word-by-word responses. Application B is a document summarizer that processes long articles and returns a complete summary after a few moments. The team needs to prioritize optimization efforts within the model's execution component, which involves an initial processing of the input followed by a step-by-step generation of the output. For which application would optimizing the step-by-step generation (decoding) stage be more critical for user satisfaction? Justify your answer by explaining how the characteristics of this stage relate to the user experience in each application.