Taxonomy of Cognitive Capability Gaps in Generative and Agentic AI

A recent study available on arXiv (2608.02553) offers a taxonomy-based examination of the cognitive capability deficiencies in generative and agentic AI. This research categorizes existing literature into five key areas: persistent state modeling, goal-directed autonomy, self-monitoring and control, interaction with environments, and learning and adaptation. The authors evaluate recent progress, pinpoint ongoing limitations, and explore unresolved research issues for each area. The objective is to advance beyond mere language generation and task execution towards systems that exhibit sustained reasoning, adaptive behavior, persistent memory, and self-regulation. Despite notable achievements in generative and agentic AI, many essential cognitive functions are still inadequately developed, hindering reliable long-term performance. The paper proposes a conceptual framework for Adaptive Cognitive Intelligence to tackle these issues, making it significant for Cognitive AI development and the wider AI research community.