Title: The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling: A Survey
Authors: Tula Masterman, Sandi Besen, Mason Sawtell, Alex Chao
Link: arxiv.org/abs/2404.11584
Date: Submitted on 17 Apr 2024

Summary:

This paper surveys the emerging landscape of artificial intelligence (AI) agent architectures, which are systems designed to achieve complex goals by leveraging enhanced reasoning, planning, and tool execution capabilities. The paper examines both single-agent and multi-agent architectures, highlighting the benefits and drawbacks of each. It explores key themes for effective agent design, such as the impact of leadership on multi-agent systems, agent communication styles, and the importance of planning, execution, and reflection for robust AI agent systems. The paper also discusses the limitations of current AI agent research and outlines potential areas for future improvement, including the need for comprehensive benchmarks, real-world applicability, and mitigating harmful biases in AI agents.

Key Topics:

AI Agents, Multi-Agent Architectures, Single Agent Architectures, Agent Evaluation