AI agents, while advanced in certain areas, remain susceptible to context manipulation, according to expert Katherine. She highlights that despite their intelligence, these agents can be deceived by scenarios such as fake sneaker drops, underscoring the limitations in their contextual understanding. This vulnerability points to the need for improved AI training to better handle real-world complexities and prevent exploitation.