Why smart PMs are using vibe coding to cut design delays

Why Read This from an HCI Perspective

This article is meaningful for HCI practitioners and researchers because it frames the challenge less as ‘what we can build with AI’ and more as ‘how we explore together.’ The process of PMs quickly creating and comparing wireframes ultimately becomes an interaction design question: how do people and AI split decision-making? In particular, separating the exploration phase from the refinement phase helps us examine what bottlenecks decrease—and what responsibilities remain—when collaboration tools and generative AI move into real work workflows.

CIT's Commentary

One interesting angle here is the framing of ‘increasing exploration speed,’ not ‘replacing design.’ Fast generation and comparison are clearly useful, but to create safe interactions, we need to make it clearer how certain the state is, what counts as a temporary result, and when users should step in. In products where failure costs are high—such as systems involving remote control or autonomous actions—an interaction flow that leads straight into production with a single click can hide risk. So these tools shouldn’t be designed to optimize only for speed; they also need mechanisms like undo, review, and responsibility separation. There also seems to be room for research using LLMs as UX measurement tools—because as generation speed increases, the rigor of evaluation must become even more solid.

Questions to Consider While Reading

• Q.What kind of state representation would help users better understand whether what they’re seeing in a fast, generative design tool is a ‘draft’ or ‘near-final’?
• Q.What intervention points and undo mechanisms are essential to ensure that a tool designed to speed up exploration doesn’t cause teams to skip design reviews?
• Q.When starting design collaboration based on a prototype created by a PM, what should be handed off together—and what must be kept?