A Stanford builder turns Apple Maps into a playable San Francisco game

Why it matters

Simonian's project shows how coding agents and map datasets can let one developer prototype a city-scale 3D experience without a conventional game studio or art pipeline.

Thijs Simonian (@cdngdev) released a browser game on August 24th that uses Apple Maps data as the raw material for a playable San Francisco, complete with climbable buildings, drivable cars and controls for roaming or flying across the city.

https://x.com/cdngdev/status/2091909073038082139

Simonian published the project in a 34-post thread on X, directing desktop users to San Francisco -- The Game. The project is an unusually large canvas for a solo web experiment: instead of modeling a few recognizable blocks, Simonian describes the map as covering the entire city.

The builder has spent years making software around San Francisco. His portfolio identifies him as a Stanford student from the city who currently works in robotics at OpenAI. It also lists earlier projects including a live map of San Francisco trains, a machine-learning parking-sign reader and a bag designed around transit data. Simonian previously worked with TreeHacks, LMArena, Andreessen Horowitz, Discord and the San Francisco Municipal Transportation Agency, according to the site.

He was TreeHacks' technology director during the Stanford hackathon's February 2026 event, where he told The Stanford Daily that organizers select for people who pursue ambitious personal projects in their spare time. The Apple Maps game applies that same builder-first approach to his own city.

A city streamed into a game

The game's desktop controls borrow from familiar open-world and sandbox conventions. Players use WASD to move, the mouse to look around, Space to jump and Shift to sprint. The C key changes the camera, H deploys a glider and V summons or enters a vehicle. Speed and viewing distance can be adjusted from the keyboard.

The live interface exposes some of the machinery underneath the experience. It reports the state of geographic tile streaming, offers detail levels labeled from Z15 through Z20 and displays a default loading range of 470 meters. New sections of the city are fetched as the player moves, turning the map into a continuous game environment rather than a fixed, handcrafted level.

That approach saves Simonian from manually rebuilding San Francisco building by building. It also creates the project's clearest constraint: the city is a large dataset that must be loaded and rendered inside a browser. Simonian told one commenter that the project involves "a lot of data." He warned phone users that mobile performance was still being optimized and recommended sending the link to a Mac.

The website displays an Apple copyright notice and identifies Apple Maps as the source of the underlying city data. Cars, movement systems and other game mechanics sit on top of that geographic layer. The result currently behaves like a technical playground rather than a structured campaign, with exploration as the central activity.

Codex meets a long-running San Francisco obsession

Simonian also pointed to AI-assisted coding in the thread. In a reply to Rajan Agarwal (@_rajanagarwal) and Kevin Zhu (@kevinbzhu), he wrote that "codex is a little too good." He did not quantify how much of the project the coding agent produced, but the comment places the game within a widening category of ambitious personal software built with agent assistance.

His public GitHub profile shows the longer track record behind the release. Simonian maintains dozens of repositories, including Skill Icons, a widely used collection of embeddable technology badges, and Truth or Dare, a Discord bot he began building before college. His portfolio says he joined Discord while still in high school and later created Video Arena, an evaluation project for generative video models.

The San Francisco game had been sitting unfinished before this week's publication. Simonian told David Lietjauw (@davidfromkansas) that seeing another map project the previous week pushed him to release his own, which had remained "on the backburner for a while."

That timing matters because the project shows what coding agents and existing map datasets can do for a single developer with a narrow idea and no traditional game studio. Simonian reused a real city as the art layer, then concentrated his work on streaming, movement and interaction. The rough edges remain visible, especially on mobile, but the project is publicly playable rather than confined to a demo video.

For Simonian, it is also a continuation of a specific body of work. His previous projects turned transit positions, parking rules and San Francisco geography into software or physical objects. This time, the city itself became the interface - and the player can press V to take a car.