From LLM Reasoning to Autonomous AI Agents: A Comprehensive Review

TL;DR

The paper surveys the rapid evolution of LLM-based autonomous agents by mapping a comprehensive benchmark taxonomy, highlighting state-of-the-art frameworks, and detailing domain-specific applications across healthcare, science, finance, and materials. It presents a side-by-side benchmark comparison (2019–2025), outlines agent protocols (ACP, MCP, A2A), and catalogs 2023–2025 agent frameworks, emphasizing multi-step reasoning, retrieval-augmented workflows, and multi-agent collaboration. Key contributions include a unified view of benchmarks, a taxonomy of ~60 evaluation suites, and a synthesis of tools, datasets, and protocols driving the field, plus recommendations on advanced reasoning, failure modes, automated discovery, dynamic tool integration, integrated search, and security. The work highlights significant performance gaps between current models and human benchmarks, underscores reliability and safety challenges, and points to practical implications for deploying autonomous AI agents in scientific research, healthcare, and industry. Overall, the survey provides a comprehensive roadmap for advancing robust, scalable, and secure agentic AI systems.

Abstract

Large language models and autonomous AI agents have evolved rapidly, resulting in a diverse array of evaluation benchmarks, frameworks, and collaboration protocols. However, the landscape remains fragmented and lacks a unified taxonomy or comprehensive survey. Therefore, we present a side-by-side comparison of benchmarks developed between 2019 and 2025 that evaluate these models and agents across multiple domains. In addition, we propose a taxonomy of approximately 60 benchmarks that cover general and academic knowledge reasoning, mathematical problem-solving, code generation and software engineering, factual grounding and retrieval, domain-specific evaluations, multimodal and embodied tasks, task orchestration, and interactive assessments. Furthermore, we review AI-agent frameworks introduced between 2023 and 2025 that integrate large language models with modular toolkits to enable autonomous decision-making and multi-step reasoning. Moreover, we present real-world applications of autonomous AI agents in materials science, biomedical research, academic ideation, software engineering, synthetic data generation, chemical reasoning, mathematical problem-solving, geographic information systems, multimedia, healthcare, and finance. We then survey key agent-to-agent collaboration protocols, namely the Agent Communication Protocol (ACP), the Model Context Protocol (MCP), and the Agent-to-Agent Protocol (A2A). Finally, we discuss recommendations for future research, focusing on advanced reasoning strategies, failure modes in multi-agent LLM systems, automated scientific discovery, dynamic tool integration via reinforcement learning, integrated search capabilities, and security vulnerabilities in agent protocols.

Figures (13)

• Figure 1: Survey Structure.
• Figure 2: Classification of LLM Benchmarks for AI Agents Applications
• Figure 3: Core Elements of AI Agents.
• Figure 4: What are Agentic Workflows?.
• Figure 5: Agent-Driven RAG Framework.