A Product Manager's guide to using AI to build working prototypes

Why Read This from an HCI Perspective

This article reframes prototyping not as a matter of producing deliverables, but as a question of how quickly you can capture user signals. For HCI practitioners, it’s a strong example of how AI tools can reduce the translation cost between design, development, and QA—and also how certain validation steps may be skipped or weakened along the way. In particular, it’s worth examining how the idea of ‘showing working software quickly’ affects interaction validation.

CIT's Commentary

An interesting point is that the focus isn’t so much on speeding up prototypes as on whether humans can clearly specify what the AI needs to respond appropriately. That, in turn, is fundamentally an interaction design problem: a good prompt leads to clear behavioral specifications, while a bad prompt results in ambiguous system states. However, once you apply this to a real product, the visibility of failure modes and the paths for user intervention become just as important as fast generation. In safety-critical contexts, ‘how quickly you built it’ is often less important than ‘when you can stop it and roll it back.’ So this isn’t merely a productivity story—it becomes a research question about how to design interactions that can be validated in the AI era.

Questions to Consider While Reading

• Q.In AI-generated interactive prototypes, at what points do users decide to trust or distrust the system?
• Q.As prototype generation speed increases, what validation gaps emerge anew across design, development, and QA?
• Q.When automating behavioral specifications or UX measurement tools using LLMs, what should be automated—and what should remain subject to human judgment?