The world of game preservation and AI-assisted development just witnessed a remarkable feat. Over a single weekend, a developer successfully resurrected a 30-year-old, text-based multiplayer game that had been considered lost to time. The tool that made this possible? Anthropic’s Claude Code.
This story, which ignited a vibrant discussion on Reddit, isn’t just about nostalgia. It’s a powerful case study demonstrating how large language models (LLMs) are transforming software archaeology and creative reconstruction. The developer, Jon Radoff (CEO of Beamable and a veteran game designer), managed to bring his 1992 creation, Legends of Future Past, back to life after it had been offline since 1999, with no surviving source code.
The Lost World of Legends of Future Past
Before World of Warcraft or EverQuest, there were MUDs (Multi-User Dungeons). These were entirely text-based virtual worlds where players typed commands like go north or attack troll to interact. Legends of Future Past was one of the earliest commercial graphical MUDs, a pioneer that won awards and built a community on early networks like CompuServe.
Jon originally developed the game over six months in 1992 on a 486 computer with just 16MB of RAM. The game world was vast, built by a team over years, featuring thousands of rooms, items, monsters, and complex systems for crafting and combat. When it shut down in 1999, it seemed gone forever—no source code, no emulators, only memories and some ancillary files remained.
The Weekend Resurrection Project
Jon’s challenge was monumental. He had:
A complete set of script files written in a custom, undocumented scripting language he invented at 19.
Some historical artifacts: a 1996 game recording, a 1998 Game Master manual, and player documentation.
What he lacked: The actual game engine source code.
He fed these materials to Claude Code with a directive: figure out what this game was and rebuild it. What followed was a tight, collaborative weekend between human and AI. As Jon described it, AI programming is not autopilot; it’s “like guiding a tireless, brilliant collaborator.” He provided context, corrected course, and made technical judgments while Claude did the heavy lifting of interpretation and code generation.
Claude’s Astonishing Technical Feat: Reverse-Engineering a “Bad” Language
The core of this achievement lies in Claude’s ability to understand and reverse-engineer a proprietary scripting language that had no formal specification. Jon himself admits the language was “pretty terrible”—a product of its time, designed to squeeze functionality into severe memory constraints. It used DOS encoding, was case-insensitive, and had idiosyncratic structures where script blocks ended implicitly.
!Original script language example
Despite this, Claude Code managed to:
- Parse the custom language from examples and a manual alone, reconstructing its interpreter.
- Decode game logic like combat formulas from GM docs and monster AI from integer-coded strategy fields.
- Understand the execution model for scripts, triggers, and in-game events.
“An AI that had never seen it before completely reconstructed the language I designed thirty years ago,” Jon remarked, clearly impressed.
From Scripts to a Live, Modern Game
Claude didn’t just understand the old scripts; it used that understanding to build a completely new, functional game system. The output was a full-stack application:
A game engine built in Go
A React-based frontend for the web
A WebSocket layer for real-time multiplayer interaction
A MongoDB database for persistence
The rebuilt world is faithful to the original, containing 2,273 rooms, 1,990 items, 297 monsters, 88 spells, and complex systems for crafting, races, and an in-game calendar. It’s now deployed and playable online.
Implications for Game Preservation and Creative Development
This project is more than a cool tech demo; it signals a shift in creative possibility. Jon contrasted the effort: in the 1990s, building such a game required months of engine work and years of world-building by a team, on expensive infrastructure. Now, the “engineering” part can be massively accelerated by AI.
“If all you need are the creative assets and a weekend, how many dead online worlds could be resurrected? How many games that only existed in someone’s imagination could now become real?”
This has profound implications:
Game Preservation: Countless classic online games from the 80s and 90s are lost. This method provides a blueprint for recovering them from partial assets.
Rapid Prototyping: Developers can focus on designing worlds, stories, and mechanics—the creative assets—while AI handles much of the underlying code translation and systems building.
Understanding Legacy Systems: Businesses struggling with outdated, undocumented “black box” systems could use similar AI-assisted analysis to understand or migrate them.
Try It Yourself and Explore the Code
The resurrected Legends of Future Past is available to play online, and Jon has open-sourced the entire project. This transparency allows other developers and enthusiasts to study the process and imagine their own applications.
Play the Game: lofp.metavert.io
Explore the Code: GitHub Repository
The Reddit thread has evolved into a celebration of early internet gaming culture, with many sharing memories of MUDs and BBS doors. Others are inspired, noting they’ve used similar AI methods to resurrect old DOS applications. This weekend project by Jon Radoff and Claude Code isn’t just about bringing back one game; it’s a glimpse into a future where the barriers between creative vision and functional reality are lower than ever, and where our digital past has a new chance at life.
Comments (0)
Log in to post a comment.
No comments yet. Be the first!