Enabling Creative Exploration for Vibe Design Agents

Vibe design agents turn natural-language briefs into rendered interfaces and frontend code. Yet a useful design agent should do more than produce one valid page: it should help users explore coherent alternatives. Increasing token-level temperature is a blunt solution because it varies aesthetic decisions and syntax-sensitive code at the same time. We instead separate exploration from implementation. Inspired by Verbalized Sampling, a pre-pass proposes several design directions with typicality scores, an external selector samples one, and the downstream generator realizes it under fixed settings. We apply this approach to UI themes and visual-asset prompts. Across 1,255 paired offline generations, candidate-level sampling broadens observed theme coverage and screenshot variation, while LLM-judge preferences vary across interventions, prompt complexity, and viewport. In an online experiment with more than 300,000 tasks, code exports increase with small latency changes, while prompt corrections also increase, revealing an exploration-steering trade-off. Together, these findings demonstrate that design-level sampling broadens the range of UI concepts explored through an explicit operational control while keeping downstream generation settings fixed.