Does who initiates a UI element matter for accountability?

Kate Moran, writing for Nielsen Norman Group, argues that "vibe coding is not generative UI (genUI)" and that "the difference is about who makes the decision to build an interactive system or component." She defines generative UI as interfaces "where the AI system decides to produce a visual or interactive element — the system initiates the design," pointing to elements like "checkboxes, interactive data visualization, clarification widgets, or full-blown complex interfaces" that the user never explicitly asked for. Vibe coding, by contrast, "means that the user describes what they want and the AI builds it" — illustrated by prompts such as "Build me a London trip planner app" or "Create a dashboard for my expenses."

The mechanism Moran proposes is that this origin point determines what the AI is held accountable for. In genUI, "that judgment call is the defining act" — the system "assessed the situation and determined that the best response would include one or more visual, interactive UI elements," which she calls "a design decision, made in real time, by the AI." In vibe coding, "the AI's role is execution — translating the user's intent into a working artifact," so the system answers for execution fidelity — "does the thing work as described?" — rather than design judgment. She notes vibe coding itself spans a range of specification, from prompts that leave the AI significant room to make structural and strategic choices to ones that leave little up to the AI system, which is why the two modes get conflated in practice.

This cuts along a different axis than Does vibe coding actually keep humans in the loop?, which places vibe coding on a human-in-the-loop spectrum relative to prompt-per-function workflows and agentic coding — a question of how much of the construction process the human stays inside. Moran's distinction instead asks who originates the decision to build a UI element at all. It also sharpens the control dimension in How should generative UI be organized as a design space? by naming the specific binary that dimension is gesturing at — AI-initiated versus user-initiated — and by tying that binary directly to an accountability split rather than leaving it as one axis among four.

Moran's framing is a conceptual distinction for practitioners, not an empirical claim tested against real systems or users; the excerpt offers no data on how often shipped products mix both modes within a single session, nor on whether users actually perceive the accountability split she describes. The implication, held to that evidentiary weight, is that evaluating an AI interface should start by asking which party initiated a given UI element before assigning credit or blame for its quality — but whether that question is answerable for any specific shipped product is left open.