Defining AI Agents: A Compendium of Criteria, Metrics, and Benchmarks

Researchers have published a new framework on arXiv that categorizes AI agents across five distinct dimensions to resolve current inconsistencies in terminology. This effort aims to standardize how developers evaluate and compare autonomous systems, providing a necessary foundation for technical reproducibility in the field. By establishing clear metrics for agentic behavior, the study addresses a growing problem where ambiguous definitions make it difficult to assess the performance and reliability of emerging AI tools.