DayNews.ai

AI agents that improve themselves hit limits

Self-improving AI agents boost benchmark scores but often lose reliability, raise costs, and fail to transfer gains to new tasks.