Jon Radoff handed Claude a collection of files from a 30-year-old game and asked it to rebuild the thing. The project - published on GitHub as "lofp" - documents how that went.
The challenge wasn't complexity in the modern sense. It was age. The files used formats from the mid-1990s: undocumented data structures, assets without surviving documentation, and logic written for hardware architectures that no longer exist. This is the kind of codebase that typically requires someone who was actually working in the field when the original was built - or weeks of manual reverse-engineering.
What Claude Was Actually Doing
Claude's role was to read file headers, infer how data was laid out, make guesses about encoding, and iteratively reconstruct working code. Radoff guided the process and made judgment calls about ambiguous choices, but the AI handled the mechanical work of interpreting undocumented formats.
This is different from the standard "write me a function" use case. It's closer to forensic analysis - looking at raw data patterns and working backward to figure out what the original programmers intended. The result is a working modern version of the game, documented step-by-step in the repository.
The Legacy Code Problem Everyone Has
Most businesses have their own version of this. Not a 30-year-old game, but a 10-year-old internal tool that nobody wants to touch because the original developer left years ago, there are no comments, and the documentation is a two-paragraph README that explains nothing.
AI coding assistants have gotten genuinely useful at this kind of archaeology. Reading unfamiliar code, inferring intent from patterns, and generating modernized equivalents is exactly the type of task where LLMs (large language models - AI systems trained on massive amounts of text and code) tend to perform well. It's not guaranteed to work, and Radoff's project required real human judgment throughout. But the ceiling is higher than most people expect.
If you have legacy code sitting untouched in your codebase, this project is a practical reference for what's now possible.