Marble Marvell is a new 3D browser game inspired by Atari’s Marble Madness

Marble Marvell starts with a simple idea. Roll a marble across a tricky 3D course, keep it on the track and reach the finish before time runs out. Anyone familiar with Atari Games’ 1984 Marble Madness will recognise the setup straight away. What makes Marble Marvell more interesting as a news story is how it was made: the browser game came together during a one-week experiment in AI-assisted development, with AI used across programming, graphics, sound and text. The result is a compact HTML5 game built with Three.js, complete with computer-controlled rivals, local two-player support and dynamic split-screen. It doesn’t try to disguise its inspiration. Marble Marvell takes the basic arcade structure of Marble Madness and rebuilds it with modern browser technology, new courses and a development process far removed from arcade production in the mid-1980s.

Marble Madness, rebuilt for a browser

The basic job is easy to explain. You steer a marble through raised courses packed with slopes, corners, gaps and moving obstacles. Momentum matters. A small mistake can send the marble over an edge, and keyboard controls give the game a different character from the trackball setup used in the arcade original.

Marble Marvell adds computer opponents and supports two players on one machine. Its split-screen system can change during play rather than keeping each player locked into a fixed half of the display. It’s a practical feature, but it’s one of the more technically involved pieces in a project built on a very short schedule.

The game was released publicly on 28 September 2026 and remains listed as a work in progress. Its scope is small, which suits the experiment. There’s no huge campaign, inventory system or story framework to manage. Most of the work sits in movement, collision, course design, opponent behaviour and presentation. Those systems still have to work together cleanly. A marble game exposes problems fast. Physics errors, awkward controls or weak collision handling are hard to hide when movement is the main attraction.

AI handled more than code

This wasn’t a case of asking a chatbot for one script and calling the project finished. AI assistance was used across several parts of production, covering code, graphics, sound and written material. A map editor was created during development too, giving the designer a way to build and adjust courses without rewriting the game every time.

Rendering work included improved shadows and performance tuning. Multiplayer added another layer, since the camera and screen layout need to react to two players moving through the same course. The project grew beyond a straight recreation exercise, with an extra campaign produced after the main mechanics, hazards, enemies and editing tools were in place.

There’s a useful distinction here. One part of the project involved giving AI detailed instructions and using its output to build systems. Another involved using existing systems to help produce fresh level layouts. Those are different jobs. Marble Marvell puts both in one small game, making it a tidy example of where AI-assisted game development currently sits: less magic button, more fast iteration with a designer directing the work.

The arcade game behind it

Marble Madness arrived in arcades in December 1984 from Atari Games. Mark Cerny designed the game, with Bob Flanagan working on programming. Players guided a marble across six isometric courses, racing a timer and dealing with ramps, drops, narrow paths, moving hazards and rival objects. A second player could join on the same course.

Its trackball control scheme gave players direct movement in any direction, a major part of the game’s identity. The cabinet ran on Atari System 1 hardware, and Brad Fuller and Hal Canon created the music using Yamaha FM audio hardware. Ports followed for home computers and consoles, helping the core idea reach players far beyond the arcade floor.

Marble Marvell doesn’t recreate the original machine at hardware level. It takes the playable concept and rebuilds it with browser tools, procedural 3D presentation and modern input. For a project produced in a week, the feature list is broad: solo play, AI rivals, local multiplayer, dynamic split-screen, custom courses and an editor.

A small release with a useful development story

As a game, Marble Marvell is direct. Roll, steer, recover, keep moving. As a development story, it’s more interesting. The project shows what one designer can assemble in a short period when generative tools take on parts of programming, asset creation and level production.

It doesn’t settle any wider debate about AI in games, nor does it try to. The useful part is simpler. Marble Marvell puts a working example in front of players, with enough systems on screen to make the experiment concrete rather than theoretical. For anyone tracking how browser games, small-scale development and generative tools are starting to overlap, this is a compact case study with a playable result.