We know that AI coding assistants excel at rapidly producing brand new applications. But what if you have a 20 year old legacy codebase, in need of extensive modernization, with hundreds of open issues that you simply don't have the time to triage, and dozens of great ideas for new features that realistically will never see the light of day? In this session, we'll explore how in my spare time, over the course of three months, armed with Claude Code, I was able to resurrect a large open source project I had thought was dead - fixing all known bugs, fully modernizing it, improving performance, and implementing multiple exciting new features. We'll focus in particular on how this experience highlighted some of the most crucial questions that the software development industry is currently wrestling with - such as "do I need to read all the code that AI is producing?", and "can I trust AI to review and test its own work?" and "is there even any point in open source software anymore if I can just ask an AI assistant to create it for me"? Most importantly, we'll ask "are the current AI tools capable of generating high-quality, production-ready code, or do they inevitably produce technical debt-ridden piles of slop"? I'll be sharing my answers to these questions, as well as the agents, skills and techniques I found most helpful in my journey to revitalize a legacy codebase.