This is cool in theory but I'm wondering if I will really want real time collaboration with an AI agent that I don't fully trust. Won't be a pain to keep track of their work and revert it?