“Scott Wu believes that improvements in AI agents will come from multiple areas simultaneously, including better foundation models, reasoning, memory, self-play, and faster inference speeds.”