I'm tempted to play around with an AI code generation "agent" like Aider or Claude Code. But that feels like eating the forbidden fruit. It's easier to stand against something on principle if you haven't already been lured in by its immediate benefits. I've already crossed that line with LLM use in general, including asking an LLM to do one-off code generation tasks (though never for my open-source work). Maybe I can at least resist the temptation to try the next level.