Skip to main content

Agentic AI Systems: Research References & Citations

Document Purpose: Comprehensive bibliography supporting the multi-planar abstraction from clinical/medical AI to general-purpose agentic systems.

Compiled: January 2026
Scope: 2023-2025 research synthesis across clinical dialogue, enterprise automation, multi-agent systems, and foundational AI architectures.

Primary Survey Sources​

Clinical Domain Origin​

  1. Zhi, Zhao, Wu, Zhao, Zhu (2025)
    "Reinventing Clinical Dialogue: Agentic Paradigms for LLM-Enabled Healthcare Communication"
    Tianjin University & Chinese Academy of Sciences
    • Contribution: Novel two-axis taxonomy (Knowledge Source × Agency Objective)
    • Scope: 300+ papers surveyed
    • Key Finding: Four paradigms emerge from intersection of implicit/explicit knowledge with cognition/execution objectives
    • Framework: POMDP formalization for clinical dialogue agents

General Agentic AI Surveys​

  1. Ali, M. A. & Dornaika, F. (2025)
    "Agentic AI: A Comprehensive Survey of Architectures, Applications, and Future Directions"
    Artificial Intelligence Review
    DOI: 10.1007/s10462-025-11422-4

    • Contribution: Dual-paradigm framework (Symbolic/Classical vs. Neural/Generative)
    • Methodology: PRISMA-based review of 90 studies (2018-2025)
    • Key Finding: "Paradigm-market fit" - symbolic dominates safety-critical, neural dominates adaptive domains
    • Recommendation: Hybrid neuro-symbolic architectures for future development

  2. MDPI Future Internet (2025)
    "The Rise of Agentic AI: A Review of Definitions, Frameworks, Architectures, Applications, Evaluation Metrics, and Challenges"
    Future Internet, 17(9), 404

    • Contribution: Comprehensive taxonomy including 143 primary studies
    • Scope: LLM-based and non-LLM agentic systems
    • Key Insight: 90%+ of reviewed papers published 2024-2025, indicating rapid field evolution

  3. Artificial Intelligence Review (2025)
    "Agentic AI systems in the age of generative models: architectures, cloud scalability, and real-world applications"
    DOI: 10.1007/s10462-025-11458-6

    • Contribution: Holistic framework integrating perception, memory, planning, execution, communication
    • Key Innovation: Persistent memory layers, semantic routing, modular orchestration

Planning & Reasoning Research​

ReAct and Derivatives​

  1. Yao, S. et al. (2024)
    "ReAct: Synergizing Reasoning and Acting in Language Models"
    ICLR 2024

    • Pattern: Thought → Action → Observation loop
    • Innovation: Interleaves reasoning traces with action execution
    • Application: Foundation for most modern agentic architectures

  2. Molinari et al. (2025)
    "RP-ReAct: A Reasoner-Planner Supervising a ReAct Executor"

    • Pattern: Hierarchical decoupling of strategic planning from execution
    • Innovation: Reasoner-Planner Agent (RPA) + Proxy Execution Agents (PEA)
    • Benefit: Prevents "contextual drift" and handles large context windows

  3. Li et al. (2024)
    "Focused ReAct: Addressing Context Drift and Action Loops"

    • Problem: Loss of original goal during extended reasoning
    • Solution: Goal reiteration and focused attention mechanisms

Reflexion & Self-Improvement​

  1. Shinn, N., Cassano, F., Labash, A. et al. (2023)
    "Reflexion: Language Agents with Verbal Reinforcement Learning"
    NeurIPS 2023
    arXiv:2303.11366

    • Innovation: Self-reflection without gradient updates
    • Components: Actor, Evaluator, Self-Reflection module, Episodic Memory
    • Result: 91% pass@1 on HumanEval (state-of-the-art at publication)
    • Key Insight: "Verbal reinforcement learning" enables learning from experience

  2. Multi-Agent Reflexion (MAR) (2024)
    "MAR: Multi-Agent Reflexion Improves Reasoning Abilities in LLMs"
    arXiv:2512.20845

    • Extension: Diverse critic personas in reflection loop
    • Framework: Society of Mind + Multi-Agent Debate (MAD)
    • Benefit: Addresses single-agent confirmation bias

Memory Systems Research​

Survey & Frameworks​

  1. ACM Transactions on Information Systems (2025)
    "A Survey on the Memory Mechanism of Large Language Model-based Agents"
    DOI: 10.1145/3748302

    • Contribution: Comprehensive taxonomy of memory types and operations
    • Framework: Parametric vs. Non-Parametric memory classification
    • Key Insight: Memory is "the key component for agent-environment interactions"
    • Repository: https://github.com/nuster1128/LLM_Agent_Memory_Survey

  2. NeurIPS 2025 (A-Mem)
    "A-Mem: Agentic Memory for LLM Agents"
    OpenReview: FiM0M8gcct

    • Innovation: Dynamic memory organization using Zettelkasten principles
    • Features: Automatic indexing, cross-referencing, interconnected knowledge networks
    • Key Advance: Memory structures adapt to task requirements

Specialized Memory Systems​

  1. Agent Workflow Memory (2024)

    • Focus: Process-oriented memory for workflow agents
    • Application: Enterprise automation, sequential tasks

  2. MemRL (2026)
    "MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory"

    • Innovation: Runtime RL on memory without model updates
    • Application: Long-horizon agent adaptation

  3. MemEvolve (2025)
    "Meta-Evolution of Agent Memory Systems"

    • Approach: Evolutionary optimization of memory architectures

Multi-Agent Systems Research​

Collaboration Mechanisms​

  1. Tran, K-T., Dao, D., Nguyen, M-D. et al. (2025)
    "Multi-Agent Collaboration Mechanisms: A Survey of LLMs"
    arXiv:2501.06322

    • Framework: Characterizes collaboration by actors, types, structures, strategies, protocols
    • Types: Cooperation, competition, coopetition
    • Topologies: Peer-to-peer, centralized (hierarchical), distributed
    • Finding: "Dominant topology in 68% of MAS implementations"

  2. ACM ICDEAI (2025)
    "A Survey of Multi-AI Agent Collaboration: Theories, Technologies and Applications"
    DOI: 10.1145/3745238.3745531

    • Scope: Cross-industry applications
    • Focus: Manufacturing, finance, healthcare coordination

  3. Multi-Agent Coordination Across Diverse Applications Survey (2025)
    arXiv:2502.14743

    • Applications: Decision-making, behavior simulation, robotics
    • Key Systems: CAMEL, MetaGPT, RoCo

Specific Multi-Agent Frameworks​

  1. Hong, S. et al. (2024)
    "MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework"
    ICLR 2024

    • Innovation: Human-like workflows in LLM-based multi-agent systems
    • Application: Software development automation

  2. AutoAgents (2024)
    "AutoAgents: A Framework for Automatic Agent Generation"
    IJCAI 2024

    • Innovation: Adaptive agent generation based on task requirements
    • Feature: Observer agent for coordination

  3. ReConcile (2024)
    "Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs"
    ACL 2024

    • Pattern: Multi-model debate with voting

Tool Use & Action Research​

  1. Stanford & Harvard (2024)
    "Adaptation of Agentic AI"

    • Framework: Agent = Foundation Model + Planning + Tool Use + Memory
    • Taxonomy: A1 (optimize agent) vs. A2 (optimize for output) vs. T1/T2 (optimize tools)
    • Key Finding: "Tool use module connects agent to external environments"

  2. τ-bench (2024)
    "A Benchmark for Tool-Agent-User Interaction in Real-World Domains"
    Yao, S. et al.
    arXiv:2406.12045

    • Contribution: Standardized evaluation for tool use

  3. ChemCrow (2023)
    Bran et al.

    • Application: 13 expert-designed chemistry tools
    • Pattern: LangChain + ReAct + domain-specific tools

Enterprise & Production Research​

  1. IBM Think (2025)
    "AI Agents in 2025: Expectations vs. Reality"

    • Prediction: AI orchestrators as backbone of enterprise AI
    • Challenge: Strong compliance frameworks needed
    • Insight: "Push and pull between multi-agent and single godlike agent"

  2. Anthropic (2024)
    "Model Context Protocol (MCP)"

    • Standard: Connecting AI assistants to data sources
    • Adoption: OpenAI support announced, thousands of integrations
    • Status: De facto standard for agent-data interactions

  3. World Journal of Advanced Research and Reviews (2025)
    "Multi-agent systems: the future of distributed AI platforms"

    • Metrics: 35% productivity gains, $2.1M annual cost reduction
    • Protocols: FIPA standards in 58% of implementations

Evaluation & Benchmarks​

  1. SWE-bench (2023)
    Jimenez, C. E. et al.
    arXiv:2310.06770

    • Focus: Real-world GitHub issue resolution

  2. GAIA (2023)
    Mialon, G. et al.

    • Focus: General AI assistant benchmark

  3. HumanEval (various)

    • Focus: Code generation evaluation
    • Significance: Standard for Reflexion and code agent evaluation

Safety & Governance Research​

  1. Khan, A. et al. (2024)
    "Governance Models for Autonomous Multi-Agent Systems"
    IEEE Access, vol. 12

    • Focus: Policy frameworks for MAS deployment

  2. CyberArk (2025)
    "The Agentic AI Revolution: 5 Unexpected Security Challenges"

    • Challenges: Identity management, secret sprawl, persistent access

  3. Prompt Injection Research (2024-2025)
    Multiple papers on LLM-to-LLM prompt injection in multi-agent systems

    • Defense: LLM Tagging and sanitization

Foundational Concepts​

LLM Agent Architecture​

  1. Lilian Weng (2023)
    "LLM Powered Autonomous Agents"
    Lil'Log
    URL: https://lilianweng.github.io/posts/2023-06-23-agent/

    • Framework: Planning + Memory + Tool Use
    • Significance: Widely-cited conceptual foundation

  2. Prompt Engineering Guide (2024-2025)
    "LLM Agents"
    URL: https://www.promptingguide.ai/research/llm-agents

    • Resource: Practical implementation guidance

Cognitive Architectures​

  1. CoALA (2024)
    "Cognitive Architectures for Language Agents"
    TMLR
    • Framework: Modular memory, action space, decision-making
    • Contribution: Organizes field and guides future development

Domain-Specific Applications​

Healthcare​

  1. daGOAT
    Autonomous GvHD prevention system

    • Paradigm: VWA (Verifiable Workflow Automator)
    • Application: Clinical protocol execution

  2. DrBioRight 2.0
    Cancer proteomics chatbot

    • Paradigm: GS (Grounded Synthesizer)
    • Application: EHR-grounded query response

  3. MedAgents / MDAgents (2024)
    NeurIPS 2024

    • Innovation: Adaptive LLM collaboration structures for medical decisions
    • Application: Collaborative diagnosis

Finance​

  1. FinRobot (2024)
    Yang, H. et al.
    arXiv:2405.14767

    • Application: Financial analysis automation
    • Framework: Open-source AI agent platform

  2. FinCon (2024)
    "LLM multi-agent system with conceptual verbal reinforcement"

    • Application: Financial decision making

Software Development​

  1. Devin (Cognition AI, 2024)
    First claimed fully autonomous AI software engineer

    • Capabilities: Shell, code editor, browser in sandbox
    • Significance: Milestone in autonomous coding agents

  2. SWE-Agent & Agentless (2024)
    Various approaches to software engineering tasks

    • Debate: Agent-based vs. simpler approaches

Research Frontiers (2025-2026)​

Neuro-Symbolic Integration​

  • Dynamic fusion of parametric/non-parametric knowledge
  • Metacognitive mechanisms (self-doubt, boundary awareness)
  • Fine-grained process supervision

Human-AI Teaming​

  • Theory of mind for agents (modeling other agents' beliefs)
  • Trust-based role arbitration
  • Consensus mechanisms for distributed decisions

Memory Evolution​

  • From RAG to true memory (non-parametric continual learning)
  • Episodic memory for long-horizon navigation
  • Privacy risks in LLM agent memory

Key Repositories & Resources​

ResourceURLPurpose
Agent Memory Surveygithub.com/nuster1128/LLM_Agent_Memory_SurveyMemory mechanisms
Awesome Agent Papersgithub.com/luo-junyu/Awesome-Agent-PapersCurated paper list
Agent Memory Paper Listgithub.com/Shichun-Liu/Agent-Memory-Paper-ListMemory-focused research
Prompt Engineering Guidepromptingguide.aiImplementation guidance
LangChainlangchain.comAgent framework
LangGraphlangchain.com/langgraphGraph-based orchestration

Citation Statistics​

CategoryPapers ReviewedDate Range
Clinical Dialogue300+2023-2025
General Agentic AI90+2018-2025
Multi-Agent Systems143+2024-2025
Memory Mechanisms100+2023-2025
Total Unique Sources400+2018-2025

Document Changelog​

VersionDateChanges
1.02026-01-14Initial compilation from survey analysis

This document supports the JSX visualizations (07-12) and markdown analyses (01-06) in the Agentic AI Systems research synthesis.