Prototyping with Prompts: Emerging Approaches and Challenges in Generative AI Design for Collaborative Software Teams

This paper looks at cross-functional teams prototyping generative AI features by iterating on prompts and examples. The content itself becomes part of the prototype, and teams have to invent new ways to collaborate when a tiny prompt change can send the model somewhere completely different. Very relatable if your day job involves making high-fidelity demos behave.
The three HCI connections I kept seeing: prompts and prompt ablations become a prototyping loop in the design life cycle; the model’s unpredictable feedback makes the gulf of evaluation feel especially narrow; and teams add constraints to prevent off-brand styles, wrong formats, and hallucinations. Apparently the prototype has a prompt menu now.
