Setting up the system with init and then fixing things manually is pretty frustrating. I’m not even convinced it works properly. It seems to pick up random strings and treat them like real people or projects. I even tried rejecting a specific proposal, but that didn’t work either. It just kept adding random words as projects or humans.
Overall, it doesn’t feel smart enough to do anything meaningful.

It would make more sense if init was only used for basic setup, like defining where the database lives. Then, the first time the system runs, the AI agent should guide the user through an interactive session. It could ask clear, step-by-step questions about projects, people, and structure, and build things up iteratively based on the user’s input. That way, the system would be grounded in real data from the start instead of guessing and getting things wrong.