Stanford now has an entire course dedicated to agents that improve themselves — covering reinforcement learning, continual learning, and the Darwin Godel Machine (which boosted its own SWE-bench score from 20% to 50% by rewriting its own code). The field is moving fast: Meta open-sourced HyperAgents, Karpathy is evangelizing autonomous research loops, and the arxiv papers are piling up. This feels like the moment self-improvement went from sci-fi premise to engineering discipline.