(1) For the Stanford AI Index, see
Stanford HAI.
(2) A research project that documents the technical and safety features of deployed AI agents. Using some 40 criteria, the 2025 AI Agent Index study evaluated more than two dozen agents. It determined there are no established standards for agents on the Web and only a small percentage disclose safety features. Most agents use GPT, Claude or Gemini models creating agent-to-model dependencies.
The AI Agent Index is available at the Cornell University Archive catalogued under Computer Science/Computers and Society (see
arXiv). See
AI agent.
How Long Do Agents Take?
The Temporal company, which offers an execution platform for building applications and workflows, defined five levels of AI agents based on time to execute, as follows:
Agent Level Time to Execute
1 - reflexive milliseconds
2 - conversational seconds to minutes
3 - orchestrated minutes to hours
4 - coordinated hours to days
5 - indefinite