In 1992, a 19-year-old built a multiplayer online game called Legends of Future Past that ran on CompuServe, won an award from Computer Gaming World, and shut down on December 31, 1999. The source code didn't survive.
What did survive: hundreds of script files written in a custom language the developer invented for Game Masters, a GM manual from 1998, and a gameplay recording from 1996. Three decades later, the developer fed all of it to Claude Code and asked it to reconstruct the game.
The result, shared this week, shows Claude Code doing something more substantive than typical code translation. The GM manual described how the scripting language behaved. The gameplay recording showed what that behavior looked like in practice. Claude Code used both as source material to reverse-engineer the system's rules, then generate working code from its own inference.
This task - reconstructing a system from documentation and output artifacts when the original source is lost - is precisely where AI assistance earns its keep. No developer would realistically spend the hours required to reconstruct a 34-year-old hobby project from first principles. Claude Code makes it feasible in an afternoon.
The broader implication: old documentation is increasingly useful as raw material for AI-assisted reconstruction. Proprietary file formats from defunct software, legacy codebases with missing context, internal tools built by employees who left years ago - Claude Code and similar tools are showing they can make meaningful progress even when the ideal starting point, complete and clean source code, isn't available.
Claude Code runs as a command-line tool that integrates AI assistance directly into your development terminal.