Legacy code, the bedrock of many critical systems, is often written in outdated languages like MUMPS and assembly, making maintenance a nightmare. Imagine trying to decipher hieroglyphics – that’s what developers face daily. But what if AI could help? New research explores how Large Language Models (LLMs) can generate documentation for these ancient systems, offering a potential lifeline for modernization efforts. Researchers experimented with four major LLMs, prompting them to create line-by-line comments for code from a real-world electronic health records system written in MUMPS and an open-source mainframe application in assembly language code. They found that, remarkably, LLMs produced surprisingly readable and accurate comments for MUMPS, often rivaling human-written documentation. However, assembly code presented a tougher challenge, with LLMs struggling to produce consistently high-quality comments. The study also highlighted a key challenge: accurately evaluating the quality of these AI-generated comments. Traditional metrics fell short, revealing a need for better ways to measure the effectiveness of LLM-driven documentation. While the quest for the perfect AI-powered documentation tool continues, this research suggests LLMs could be a powerful ally in taming the legacy code beast, potentially saving organizations time, money, and a few headaches along the way.