# The AI Agent Index

**Authors:** Stephen Casper, Luke Bailey, Rosco Hunter, Carson Ezell, Emma Cabalé, Michael Gerovitch, Stewart Slocum, Kevin Wei, Nikola Jurkovic, Ariba Khan, Phillip J. K. Christoffersen, A. Pinar Ozisik, Rakshit Trivedi, Dylan Hadfield-Menell, Noam Kolt  
**arXiv:** [2502.01635](https://arxiv.org/abs/2502.01635)  
**Date:** 2025-02-03  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2502.01635v1/

## Key points

- The rise of agentic AI systems in the rapidly evolving landscape of artificial intelligence
- Lack of structured framework for documenting technical components, intended uses, and safety features of agentic systems
- Introduction of the AI Agent Index as the first public database to compile information on deployed agentic AI systems
- Details cataloged in the index include system components, application domains, and risk management practices
- Notable lack of transparency regarding safety measures despite ample information on capabilities and applications
- Increase in deployments of agentic systems since 2023 with a significant uptick in the latter half of 2024
- Majority of indexed systems originate from developers in the USA specializing primarily in software engineering and computer use
- Concerns about potential problematic practices due to insufficient transparency around safety features

## Summary

The rapidly evolving landscape of artificial intelligence (AI) has seen a rise in the deployment of agentic AI systems. These systems are capable of planning and executing complex tasks with minimal human intervention, making them highly sought after by developers and startups. However, there is currently a lack of structured framework for documenting the technical components, intended uses, and safety features of these agentic systems. To address this gap, the AI Agent Index has been introduced as the first public database to compile information on deployed agentic AI systems. This index catalogs details such as system components (including base model, reasoning implementation, tool use), application domains (such as computer use and software engineering), and risk management practices (evaluation results, guardrails). While developers provide ample information on capabilities and applications of agentic systems, there is a notable lack of transparency regarding safety measures. The index reveals that agentic systems have been increasingly deployed since 2023, with a significant uptick in deployments in the latter half of 2024. The majority of indexed systems originate from developers in the USA, with a mix of academic and industry-based projects specializing primarily in software engineering and computer use. These systems offer a glimpse into the emerging field of AI agents. However, concerns arise about potential problematic practices stemming from insufficient transparency around safety features. Developers may manipulate information disclosure to present a favorable image without addressing critical risk management aspects.