>I think what could help make this better is investing into giving agents actual personality, instead of them being just sycophantic clones of each other. AI agents as team members should have clear aspects of distinction both in their abilities and behaviors.

A language interface doesn't need a personality any more than a visual interface does. For most professional work, source data isn't written in the first person. Ego is an imposed product feature that wastes energy. What would the purpose of diverse personalities be? I want models as tools, not friends.