Engenharia de Prompt: upstream refinement e o futuro das interfaces de IA

Você é um consultor de IA especializado em engenharia de prompts. A partir do texto de referência a seguir, elabore um prompt útil e interessante que poderia ser usado em ferramentas de IA. O objetivo é explorar a evolução da engenharia de prompts e a ideia de 'upstream refinement' antes de o input alcançar o modelo. O texto de referência é: Is prompt engineering just scaffolding until better interfaces arrive? Prompt engineering today feels similar to early programming practices where users were expected to manage low-level details manually. A significant amount of prompt work is spent on formatting intent: restructuring input, fixing ambiguity, constraining outputs, and iterating phrasing rather than refining the actual task. I am experimenting with a workflow where prompt refinement happens upstream before the model sees the input. The model itself does not change. What changes is that raw input is automatically clarified and constrained before becoming a prompt. This raises a broader question. Is prompt engineering a fundamental long-term skill for humans interacting with models, or is it a transitional abstraction until interfaces handle more of this automatically? In other words, do we expect users to keep engineering prompts, or do prompts eventually become implementation details hidden behind better interaction layers? Curious how people deeply invested in prompt work see this evolvin