Better AI Without a Better Model

AI agents are getting smarter without needing better underlying models, by using memory, reflection, and self-rewriting code. Systems like Voyager, Reflexion, and Google's AlphaEvolve showed that agents can accumulate experience and improve through structured learning loops. The Darwin Gödel Machine took it further by rewriting its own code, boosting benchmark scores dramatically. Today's key insight is that smart harness design now matters as much as model quality.