Tag
The article discusses what data agents should save for safe replay of failed runs, distinguishing between reproducing failures and resuming after fixes, while balancing automatic retries with human intervention.
This paper introduces operational resilience and considerate participation as metrics for evaluating generative AI agents in sustained workflows, studying their behavior under increasing challenges in simulated healthcare trajectories.