A new tool called Unlegacy wants to solve a problem most companies would rather ignore: documenting code that spans decades. The project, posted to Hacker News this week, handles everything from COBOL running on mainframes to code spat out by AI coding assistants.

Most documentation tools pick a lane. They're either for modern stacks or legacy systems. Unlegacy is betting that enterprises need both in one place.

And there's real demand for this. Banks and government agencies still run on COBOL systems that predate most of their current developers. These same organizations are now experimenting with AI code, creating a weird collision of old and new. Understanding what legacy systems actually do before you replace them is hard, expensive work. AI-generated code introduces its own documentation gaps, similar to the struggles of AI documentation assistants.

The technical details are thin. Unlegacy's public materials don't explain exactly how it parses different languages or what the output looks like. Without seeing concrete documentation examples or benchmarks, it's hard to assess how well it delivers on the promise.

That's the real question. Can one tool document COBOL and AI code equally well, or does trying to be universal mean being mediocre at everything?