Yup. Codebase structure for agents is a rabbit hole I've spent a lot of time going down. The interesting thing is that it's mostly the same structure that humans tend to prefer, with a few tweaks: agents like smaller files/functions (more precise reads/edits), strongly typed functional programming, doc-comments with examples and hyperlinks to additional context, smaller directories with semantic subgroups, long/distinct variable names, etc.