Prompt Design

Why Read This from an HCI Perspective

This article reframes prompting not as simple sentence writing, but as a design practice for handling input variables. For HCI/UX practitioners and researchers, it helps them view LLM interactions as a problem similar to screen design, and encourages thinking about how established interface principles—such as readability, composability, and responsiveness—can be translated to AI interfaces. In particular, the perspective that treats prompts as a ‘renderable artifact’ is especially useful in practice.

CIT's Commentary

From a CIT perspective, the core contribution of this article is that it shifts prompting from language use to interface design. The analogy to web design may feel somewhat exaggerated, but the suggestion to actually render and review ‘invisible internal instructions’ is highly important in HCI. Concepts such as JSX-based compositional authoring, real-data previews, and partial re-rendering turn prompts from something shaped by an individual developer’s intuition into a repeatable design workflow. However, because model performance and API constraints change rapidly, an additional framework is needed to distinguish design principles from runtime constraints.

Questions to Consider While Reading

• Q.If we treat prompts as ‘renderable interfaces’ like web pages, what interaction-quality aspects should HCI evaluation metrics include beyond accuracy?
• Q.How much does JSX-based prompt design actually help with collaboration and maintenance, and can we conduct empirical studies comparing it with existing string-template approaches?
• Q.In an environment where models and APIs change quickly, what stable design principles does ‘prompt design’ have, and when should abstraction move into a framework or compiler layer?