Swiss Developer Uses 15 AI Agents to Build a Live 3D Map of the Canton of Geneva
Swiss developer Arnaud Ricci has built Geneva 3D, a browser-based three-dimensional live map of the canton of Geneva, by directing up to 15 AI agents rather than writing the code by hand, Digitec’s online magazine reported on October 21, citing an account in ICTjournal. The project started in July 2026 as a plan to show the Geneva public transport network and kept growing.
The map draws on publicly available data from more than 15 institutions, including MeteoSwiss, other federal agencies and NASA. Users can switch on layers for vegetation, street lights, a real-time day-night cycle, a weather layer that shows raindrops when it rains and current temperatures from measuring stations. Vehicle positions are partly calculated from timetables and real-time information rather than GPS, and road traffic is simulated from official counting stations, with fewer lorries at night and on public holidays. A reconstruction of Geneva around 1850 can also be displayed.
Ricci talked to one main agent, mostly running on Anthropic’s Claude models, which delegated tasks to the others. The agents did most of the routine work while he concentrated on architecture, coordination and quality control. He estimates the project would have taken two to three years without agentic help, and says the cost of the agents stayed modest despite heavy use. Geneva 3D, built on the Three.js JavaScript library and rendered on the user’s own device, is to remain free.
What it means in practice
The limits are as instructive as the speed. The agents struggled most with visuals, which they cannot judge directly; Ricci had them take and analyse screenshots, an approach that partly broke down for real-time animations, and shadows on 3D objects needed manual rework. The project also needed several code restructurings as features were added. That pattern of fast generation followed by human checking is the problem Zurich’s Klarent is tackling with deterministic AI software testing.
The project also shows what open public data can do in a small country. The same kind of timetable and traffic feeds underpin more serious planning work, as in our look at Alpine corridor slotting, and Bern is weighing five priority areas for AI in the federal administration.
What to watch: whether Geneva’s public bodies take an interest in the map, whether similar agent-built tools appear for other cantons, and how such projects make clear where simulated vehicles sit alongside real measurements.