Here’s my submission for the Google Stitch Challenge A vibe mapping product called Drift! It started with a small idea: what if a map could feel like the place it shows?

What is it and what does it do?

DRIFT turns every neighbourhood into its own visual identity. As a p0, starting with London. Camden is a punk zine. Chelsea is an oil painting. Soho is manga. Brixton pulses with Caribbean folk art. 93 neighbourhoods, 12+ art styles. All of these were derived from real Google Maps Platform data. The art style IS the information. APIs: Maps JavaScript API, Places API (New), Geocoding API.

How I Built It?

Every screen was built through Stitch’s streaming agent by describing what I wanted, watching it generate live, correcting through the agent. For all screens, I used variants to explore different art directions. DESIGN.md captured the emergent design system; I exported and re-imported it for consistency. I then prototyped in Play Mode, then exported to AI Studio and deployed on Cloud Run. Features used: Streaming Agent, Agent Manager, Variants, DESIGN.md Export/Import, Play Mode, Code Export.

Stitch Feedback

What I loved: Image quality is exceptional. I almost want to print and frame some the art it generated. Variants are absolutely essential for creative work and generating variants once saved me a lot of back and forth. The workspace itself is very smooth with low latency. Fantastic selection of export integrations available, with key platforms covered.

What could be made better: Context doesn’t persist between screens (need a “project rules” feature). Undo wipes entire history, not just the last change. Chats are not threaded in the same way as other genAI chatbots, which took a bit of getting used to. Exports to build envirenments outside of AI Studio is restrictive (I tried Lovable as well and could only export 10 screens).