Jul 2026· WIREs Data Mining and Knowledge Discovery· Vol 16· 1 citation· 62 references
TL;DR
An organizing framework for understanding LLM‐based agents is established, systematically deconstructing both single‐agent and multi‐agent systems into their core components, and the architectural principles and key mechanisms that underpin their intelligence are analyzed.
Abstract
The concept of the intelligent agent represents a long‐standing pursuit in artificial intelligence. Recent breakthroughs in large language models (LLMs) have catalyzed a paradigm shift, enabling the development of sophisticated agents that exhibit advanced reasoning, planning, and tool‐use capabilities across diverse domains. These LLM‐based agents, which leverage natural language as a universal interface for cognition and interaction, are rapidly advancing from theoretical constructs to practical applications, ranging from autonomous task assistants to complex multi‐agent simulations of social and economic systems. This paper provides an integrative survey of this burgeoning field. We first establish an organizing framework for understanding LLM‐based agents, systematically deconstructing both single‐agent and multi‐agent systems into their core components. We analyze the architectural principles and key mechanisms that underpin their intelligence, including planning paradigms, memory structures, and reflection‐based self‐improvement. We further investigate the dynamics of multi‐agent systems, exploring coordination strategies, communication protocols, and organizational structures. The paper also covers the crucial aspects of performance evaluation, highlighting influential benchmarks and identifying key challenges. Finally, we synthesize the current landscape to discuss the primary challenges, such as the intrinsic limitations of LLMs and the complexities of ensuring safety and alignment, and chart a course for future research directions, including the drive toward continual learning and enhanced multi‐modal capabilities.
Large Language Model (LLM)–based agents are rapidly evolving from passive assistants into autonomous, tool-using, and collaborative systems capable of executing complex, long-horizon tasks across web, software, and physical environments. However, the current literature remains fragmented, with inconsistent terminology, ad hoc architectures, and limited evaluation standards, making it difficult to compare systems or deploy them reliably in real-world settings. This paper presents a unified, taxonomy-driven, and deployment-oriented survey of agentic AI systems, synthesizing recent advances through a modular reference architecture and a four-dimensional taxonomy that characterizes agents along the axes of autonomy, tool use, collaboration, and safety–governance. We systematically analyze representative single-agent, tool-augmented, and multi-agent frameworks within this taxonomy, highlighting design trade-offs, capability scaling patterns, and recurring failure modes. Beyond architectural analysis, we review emerging evaluation methodologies that move beyond static benchmarks to assess agent behavior, robustness, grounding, and operational cost in interactive environments. Importantly, the survey emphasizes practical considerations for enterprise and safety-critical deployment, including access control, human-in-the-loop oversight, and policy enforcement. By unifying conceptual foundations with empirical trends and deployment constraints, this work provides a structured roadmap for researchers and practitioners to design, evaluate, and govern next-generation LLM-based agentic systems.
Sparsh Bajoria, Shreyanshu Ranjan, Adhitya M et al.· Cognitive Computation· 0 citations
The increasing availability of Large Language Model (LLM)-driven agents has introduced a new class of Artificial Intelligence (AI) systems capable of processing, organizing, and interacting with heterogeneous data sources using natural language. Although these systems are being adopted across various industries, their application in mining, particularly within Dynamic Mine Planning (DMP), remains nascent. A key research gap is the limited understanding of how LLM-driven agents can support multisource data integration across geological, operational, and safety domains in DMP to support adaptive decision-making. Traditional planning pipelines are typically too static to accommodate operational variability and real-time responsiveness. To address this gap, the review is structured around five research questions: agent applications in mining, transferable architectures from related domains, adoption barriers, multi-agent opportunities for real-time decision support, and future deployment priorities. Through a PRISMA-based synthesis of 33 peer-reviewed studies across mining and adjacent engineered sectors, the review examines LLM agents within the broader landscape of AI-embedded decision-making systems, from dispatch automation and predictive maintenance to Mining 5.0 frameworks, identifying current capabilities, performance benchmarks, and methodological gaps. The analysis proposes an incremental maturity pathway, advancing from bounded advisory systems to fully integrated planning frameworks tailored to mining’s operational requirements. As mining-specific LLM-agent implementations are largely conceptual, the pathway serves as a theoretically grounded framework to guide future empirical validation in mining. The proposed pathway offers a structured, conceptual reference for phasing in LLM-driven decision support toward greater autonomy. The findings advance both academic research and industry practice by clarifying the necessary data infrastructure, hybrid architectures, safety validation mechanisms, and governance conditions for deployment in mining operations.
Ricardo Nunes, Nathalie Risso, M. Momayez· IEEE Access· 0 citations
The rapid advancement of Large Language Models (LLMs) has significantly expanded the capabilities of Artificial Intelligence in language understanding, reasoning, and automated decision support. Despite these achievements, systems built around a single language model remain vulnerable to problems such as factual inaccuracies, hallucinated information, inconsistent outputs, limited explainability, and unintended bias. These shortcomings restrict their use in applications where decisions must be accurate, transparent, and accountable. This work introduces the Language Model Council (LMC), a collaborative framework that combines the expertise of multiple specialized AI agents to evaluate a user query from different perspectives. Their independent analyses are consolidated through a consensus-driven mechanism that selects the most reliable response. To further improve transparency, the framework integrates Explainable Artificial Intelligence (XAI), providing confidence estimates together with concise reasoning summaries that clarify how the final decision was derived. Experimental evaluation indicates that the proposed approach outperforms traditional single-model systems by improving response quality, reducing hallucinations, and increasing user trust through enhanced explainability.
Deekshitha M, Shwetha KR, Divya G S et al.· International Research Journ...· 0 citations
Large Language Models have rapidly advanced, enabling composite multi-agent systems capable of sophisticated text-based collaboration. In domains that demand coordinated reasoning, teams of agents can combine complementary strengths to produce better results than a single model. In practice, multi-agent decision making faces recurring shortcomings: agents frequently operate with poorly grounded or outdated knowledge, they seldom represent peers’ beliefs and intentions in a disciplined manner, and their conclusions are rarely checked against explicit logical constraints. These gaps reduce coherence, hinder dependable cooperation, and make the reasoning chain difficult to explore, often manifesting their black-box nature. How can a practical MAS architecture be designed to ground agents with relevant information, support explicit modelling of collaborators’ beliefs, and validate outputs against formal rules? How can the individual and synergized contributions of these mechanisms be measured across multiple case studies? To explore these questions, we present a purpose-built experimental platform. The system integrates a retrieval-augmented generation pipeline to ground agents in relevant materials, a Theory of Mind module that makes agent beliefs explicit, and an LLM-based logic auditor that evaluates outputs against an explicit rule set. Applied to diverse text-based decision-support and coordination case studies, the platform and protocol provide a reproducible blueprint for investigating how grounding, social reasoning, and logic validation together improve multi-agent collaboration. In doing so, this study contributes to the field of artificial intelligence by offering a structured approach to building, testing, and refining multi-agent architectures that balance knowledge grounding, perspective modelling, and reasoning validation.
Large Language Models (LLMs) are evolving from performing end-to-end mathematical reasoning to integrating agentic intelligence. However, most existing math benchmarks evaluate only final answers. This outcome-oriented evaluation provides limited diagnostic value for identifying process-level failures or rigorous logic, failing to guide the transformation of LLMs into robust agents. To bridge this gap, we present a process-level benchmark designed to evaluate the inherent agentic mathematical reasoning abilities of LLMs. Our framework aligns problem-solving agentic behaviors with a structured taxonomy of reusable mathematical atomic capabilities. We design a comprehensive suite of planning, action, and feedback tasks across both textual and multimodal contexts, supported by an automated pipeline that synthesizes high-quality trajectories and produces fine-grained annotations via controlled LLM rewriting. Experiments reveal that models with similar end-to-end accuracy can exhibit markedly different agentic capability profiles. This demonstrates that process-level evaluation is crucial for interpreting the true potential of LLMs and guiding the development of next-generation mathematical agents.
Jiayi Kuang, Yinghui Li, Yun-Ze Song et al.· 0 citations
Large Language Model (LLM)-based agents are evolving from isolated task executors into interconnected societies of autonomous services capable of coordination, adaptation, and collective intelligence. This paper surveys and synthesizes recent advances in agentic services computing, LLM-based multiagent systems, and language-augmented reinforcement learning to analyze how feedback-driven learning loops enable emergent behaviors at system scale. We organize the design space along four dimensions: perception and context modeling, autonomous decision-making, multi-agent collaboration, and evaluation with alignment and trustworthiness. Building on this analysis, we propose a reference architecture for feedback-driven LLM-agent societies that integrates reinforcement learning, verbal feedback, episodic memory, coordination, and governance layers. We further define sociocognitive execution metrics for coordination density, goal agreement, role specialization, recovery, strategy diversity, throughput, behavioral variance, and failure tolerance, and illustrate their use through case studies and a localized ASC Micro-Testbed prototype. The prototype results show how critic feedback, episodic memory, and macro-level safety filtering support bounded recovery and constraint preservation. Finally, we identify open challenges, including cumulative learning without knowledge entropy, scalable coordination, trustworthy evolution, and standardized evaluation for reliable emergent agentic systems.
Sadaf Shafi, Michael Bidollahkhani, Julian M. Kunkel· Annual International Compute...· 0 citations