Thirty years after a text-based multiplayer game went dark, a developer brought it back without a single line of original source code.
The project centered on a 1992 MUD - short for Multi-User Dungeon, a type of text-based online game that predates modern MMOs by decades. When the source code was lost, what remained was what typically survives: binary executables, documentation fragments, saved game data, and in some cases the logged memories of players who put hundreds of hours into those worlds.
The reconstruction used agentic AI - systems that plan and execute multi-step tasks independently, rather than just answering a single question - to analyze the artifacts, infer the game's underlying logic, and piece together a working version. The AI acted less like a code generator and more like a forensic archaeologist, deducing rules from observable behavior rather than reading them from source.
This matters beyond nostalgia. Game preservation has long been a lossy process. Emulation handles hardware differences, but when source code is gone entirely, you're typically stuck with whatever executable still runs on modern hardware. AI-assisted reconstruction offers a different path: extract intent from behavior, then rebuild.
The approach has real limits. A game reconstructed from artifacts reflects what was observable, not necessarily what was intended. Bugs from the original might become permanent features if the AI infers them as deliberate logic. But for communities that have lost their software heritage, a working approximation is better than nothing - and this project is an early proof that the technique is viable.