Continual learning and evaluating how AI agents learn across sequences of tasks

Examines how agent systems retain and apply experience across related tasks rather than treating each independently, introducing a “gain” metric to isolate learning from baseline capability.