The Evolution of Autonomous AI Platforms Integrating LLM Inference, Multi-Agent Reinforcement Learning, and Self-Healing Architectures

Abstract

Enabling adaptive decision-making, learning, and autonomous behavior of modern intelligent systems with minimal human intervention. Recent developments in Large Language Models (LLMs), Multi-Agent Reinforcement Learning (MARL), and self-healing architectures have greatly improved the capabilities of autonomous systems in complex and distributed environments. The survey will review the history of autonomous AI systems and discuss the use of LLMs as cognitive engines for reasoning, planning, and natural language dialogue. Also, it covers the methods of MARL and how they allow multiple intelligent agents to cooperate and compete. Moreover, self-healing architectures studied, showing their contribution to improving resilience of the systems via fault tolerance, automation of restoration, and prediction-based maintenance. Different deployment models, architectural approaches, applications, and performance measures will be considered to give a complete picture of contemporary autonomous AI systems. Additionally, important problems such as scalability, reliability, security, privacy, explain ability, and resource limitations will be discussed. Emerging trends of agentic AI systems, autonomous AI operating systems, foundational models for MARL, and architectures inspired by AGI will also be presented. This survey will present information about the convergence of LLMs, MARL, and self-healing approaches and possible future research directions in the field of autonomous AI systems.