We never really optimized @convex for agents, but it turns out agents benefit from the same architectural simplifications humans do. Difference is humans take years to adopt something better even when it's obviously better. Agents do what you tell them immediately. Having no ego or bias is perhaps their strongest suit.