Nubank’s Flutter app serves more than 125 million customers across Brazil, Mexico, and Colombia. At that scale, consistent delivery requires canonical approaches: Nu Design System standardizes the UI, while Backend Driven Content enables hundreds of teams to ship experiences through a single multi-country application. But production delivery and exploration optimize for different things. The canonical path provides safety, consistency, and governance; exploration needs variation, autonomy, and a low cost of being wrong. Nubank’s Concept Car was created as a structured, non-production Flutter environment where designers and engineers can stress-test motion, transitions, loading states, haptics, sound, and the boundaries of Flutter itself — without bypassing the discipline of the production ecosystem. AI is now changing who can participate in that process. Designers can move from design intent toward runnable Flutter experiments more independently, shortening the distance between an idea and something that can be experienced. But once creating experiments becomes faster, distribution becomes the next bottleneck: how can another designer or stakeholder open the result on a real device without waiting for packaging, CI/CD, or traditional app distribution? In this joint talk, Nubank and LeanCode share how Concept Car provides the environment, AI gives designers greater agency, and Flitz gives experiments reach. We’ll show how Flutter preview environments can deliver portable bytecode bundles to real devices in minutes, while keeping the production release pipeline intact — and discuss the trade-offs of building a fast feedback lane alongside a mobile platform operating at massive scale.