In this podcast episode, Andrej Karpathy discusses why he believes we're in the "decade of agents" rather than the "year of agents," explaining that current AI agents lack key capabilities like continual learning and multimodality. He critiques reinforcement learning as "terrible," discusses why model collapse prevents LLMs from learning like humans, predicts AGI will gradually blend into existing economic growth rather than cause sudden disruption, explains why self-driving cars took so long to develop, and shares his vision for the future of education.