305 Papers on Zone III Agentic AI

Tool use enables large language models (LLMs) to access external information, invoke software systems, and act in digital environments beyond what can be solved from model parameters alone. Early research mainly studied …

Large language models are increasingly acting as epistemic agents, influencing our knowledge environment and decision-making. This paper argues that the impact of these AI agents on knowledge creation and synthesis neces…

This survey presents a structured and comprehensive analysis of evaluation methodologies for Agentic AI, introducing an eleven-dimensional taxonomy. It systematically examines benchmarks, frameworks, and evaluation tools…

This study synthesizes current knowledge on Agentic AI (AAI) by examining its conceptual foundations, practical applications, and emerging research directions. It distinguishes AAI from traditional automation and generat…

This survey synthesizes the emerging landscape of AI agent architectures, covering deliberation, reasoning, planning, control, tool calling, and environment interaction. It organizes prior work into a unified taxonomy sp…

This paper argues that log analysis is crucial for credible evaluation of AI agents, as outcome-based benchmarks alone can be misleading. It presents a taxonomy of threats to credible evaluation, including inflated/defla…

This empirical study investigates the testing practices employed in open-source AI agent frameworks and agentic applications. It analyzes a dataset of 107 agent framework repositories to identify common testing strategie…

Recent advances in large language models (LLMs) provide powerful capabilities for knowledge-driven course planning. However, building reliable, constraint-aware study planners from publicly available course webpages rema…

Although the latest artificial intelligence technologies can greatly improve work efficiency by automatically generating feasible solutions in the digital world (DW), they are incapable of discovering or creating new kno…

The risks posed by AI features are increasing as they are rapidly integrated into software applications. In this paper, we present an agentic framework that constructs knowledge graphs (KGs) from AI policy documents and …

Generative AI, particularly Large Language Models, increasingly integrates graph-based representations to enhance reasoning, retrieval, and structured decision-making. This survey provides a concise, structured overview …

We introduce ResearchGym, a benchmark and execution environment for evaluating AI agents on end-to-end research. In a controlled evaluation of an agent powered by GPT-5, we observe a sharp capability--reliability gap. Re…

Large Language Models (LLMs) promise to accelerate discovery by reasoning across the expanding scientific landscape. We introduce a multi-agent framework guided by large-scale knowledge graphs to find sustainable substit…

With the recent advancements in Large Language Models (LLMs) and Agentic AI, Context Engineering (CE) has emerged as a novel research area. Knowledge Graphs (KGs) offer a promising approach to integrate diverse contextua…

This paper provides an anatomical review of efficient large language model (LLM)-based agents, focusing on memory, tool learning, and planning mechanisms. It systematizes diverse approaches from 2023 to 2025, highlightin…

Agentic AI workflows often suffer from high error rates in multi-step processes, hindering their reliability in critical applications. This paper introduces GraphFlow, an architecture for formally verifiable visual workf…

Agentic AI workflows offer significant potential for automation, but their scalability and computational cost remain critical challenges. This paper presents AAFLOW, a framework of scalable patterns designed to optimize …

Agentic AI enables large language models (LLMs) to dynamically reason, plan, and interact with tools to solve complex tasks. However, agentic workflows often suffer from inefficiencies due to suboptimal tool utilization …

Artificial Intelligence is moving from models that only generate text to Agentic AI, where systems behave as autonomous entities that can perceive, reason, plan, and act. This paper investigates architectures and propose…

Multi-turn agent systems based on Large Language Models (LLMs) face significant efficiency concerns due to the high computational cost of input tokens from ever-growing trajectories. This paper introduces AgentDiet, an i…

The 2026 Deloitte AI report highlights that while agentic AI usage is poised to rise sharply, oversight is lagging, with only one in five companies having a mature governance model for autonomous AI agents. The report co…

The rapid proliferation of agentic artificial intelligence (AI) systems, which are autonomous agents capable of perceiving, reasoning, planning, and executing multi-step tasks with minimal human intervention, presents fo…

AI systems that use reasoning to autonomously pursue goals through interaction with external environments and tools—referred to hereafter as “AI agents” or “agentic AI”—promise transformative benefits for productivity an…

LLM-based agents struggle with long-horizon planning due to losing track of goals and sparse rewards in RL fine-tuning. This paper proposes a subgoal-driven framework with an agent that leverages proprietary models for o…

LLM-based agents often fail to sustain coherent behavior over long planning horizons due to a mismatch between step-wise reasoning and long-horizon planning. This paper argues that locally optimal choices lead to myopic …

Long-horizon task planning for robotic autonomy often results in logically inconsistent or physically infeasible plans from LLM-based agents. This paper proposes LLM-KGPlan, a systematic framework combining rule-guided C…

Agentic AI represents a transformative shift in artificial intelligence, but its rapid advancement has led to a fragmented understanding, often conflating modern neural systems with outdated symbolic models—a practice kn…

Large Language Models (LLMs) are increasingly deployed as agentic systems that plan, memorize, and act in open-world environments. This shift brings new security problems: failures are no longer only unsafe text generati…

AI agents that combine large language models with non-AI system components are rapidly emerging in real-world applications, offering unprecedented automation and flexibility. However, this unprecedented flexibility intro…

Agentic artificial intelligence (AI) systems are expected to have transformative impacts across sectors, including critical areas like finance and healthcare. Their architectural complexity, autonomous decision-making ab…

Adversarial robustness in artificial intelligence is commonly defined in terms of input-level perturbations applied to static models. This study reconceptualises adversarial vulnerability for artificial and agentic AI sy…

Large Language Models (LLMs) suffer from factual hallucinations, meaning the LLMs confidently provide responses that are inconsistent with reality. Previous studies explored fine-tuning-based verbalized confidence calibr…

Large Language Models (LLMs) have demonstrated remarkable proficiency in natural language processing applications, encompassing question answering, text generation, and reasoning capabilities. However, their metacognitiv…

Large Language Models (LLMs) often produce incorrect answers on multi-hop question answering even when the reasoning trace already contains a correct intermediate conclusion. We attribute this gap to weak self-regulation…

While Chain-of-Thought (CoT) prompting advances LLM reasoning, challenges persist in consistency, accuracy, and self-correction, especially for complex or ethically sensitive tasks. Existing single-dimensional reflection…

AI Agents have rapidly gained prominence in both research and industry as systems that extend large language models with planning, tool use, memory, and goal-directed action. Despite this progress, the development and ma…

This study addresses the application of Artificial Intelligence (AI) agents to enhance national security against Advanced Persistent Threats (APTs) in cyber warfare. With the increasing sophistication of cyber threats, p…

Agentic Artificial Intelligence (Agentic AI) is emerging as a practical paradigm for coordinating autonomous decision workflows in industrial asset management. This paper proposes an event-driven multi-agent architecture…

Traditional AI coding assistants operate as single agents responding to immediate user requests, lacking persistence, organizational structure, and the ability to coordinate complex, long-running tasks. Existing multi-ag…

The rapid advancement of large language models (LLMs) has catalyzed a transition from conventional, manually orchestrated data science workflows toward autonomous analytical systems capable of iterative reasoning, tool i…

This paper argues that existing benchmarks for LLM agents fail to meet the stringent legal, procedural, and structural requirements of the public sector. It defines criteria for public sector-relevant benchmarks, includi…

Large Language Models (LLMs) have demonstrated remarkable capabilities in various natural language processing tasks, including text summarization. However, a significant challenge in deploying LLMs for critical applicati…

While agent evaluation has shifted toward long-horizon tasks, most benchmarks still emphasize local, step-level reasoning rather than the global constrained optimization (e.g., time and financial budgets) that demands ge…

Multi-agent systems (MASs) have emerged as a promising paradigm for automated code generation, demonstrating impressive performance on established benchmarks. Despite their prosperous development, the fundamental mechani…

Does continued scaling of large language models (LLMs) yield diminishing returns? In this work, we show that short-task benchmarks may give an illusion of slowing progress, as even marginal gains in single-step accuracy …

The increasing complexity of user demands necessitates automation frameworks that can reliably translate open-ended instructions into robust, multi-step workflows. Current monolithic agent architectures often struggle wi…

Large Language Models (LLMs) have advanced artificial intelligence by enabling human-like text generation and natural language understanding. However, their reliance on static training data limits their ability to respon…

Retrieval-Augmented Generation (RAG) systems are increasingly evolving into agentic architectures where large language models autonomously coordinate multi-step reasoning, dynamic memory management, and iterative retriev…

This paper proposes Contextual Memory Virtualisation (CMV) to abstract away strict physical token limits of the LLM context window. It uses DAG-based state management and structurally lossless trimming to manage memory s…

This paper introduces M4, a meta-memory framework for LLMs to evaluate its effectiveness in memory management. It features active memory management to prevent unbounded context growth, bounded by finite memory chains.…

Retrieval-Augmented Generation (RAG) systems are usually defined by the combination of a generator and a retrieval component that extracts textual context from a knowledge base to answer user queries. However, such basic…

This work presents ReMemR1, which integrates memory retrieval into the memory update process, enabling the agent to selectively callback historical memories for non-linear reasoning. It also proposes a multi-level reward…

This paper compares a fact-based memory system built on the Mem0 framework against long-context LLM inference on three memory-centric benchmarks. It evaluates both architectures on accuracy and cumulative API cost, showi…

Memory emerges as the core module in the large language model (LLM)-based agents for long-horizon complex tasks. This paper summarizes a unified framework that incorporates all existing agent memory methods and extensive…

In the era of the intelligent science paradigm, AI governance has emerged as a focal point of global attention. This study systematically reviews the current status of global AI governance research, which contributes to …

Financial services institutions are deploying sophisticated GenAI and agentic AI systems within governance frameworks designed for traditional statistical models, creating a widening gap between AI risk and institutional…

Advancements in model algorithms, the growth of foundational models, and access to high-quality datasets have propelled the evolution of Artificial Intelligence Generated Content (AIGC). Despite its notable successes, AI…

Artificial intelligence (AI) systems are moving from assistive tools into enterprise workflows where software agents retrieve information, call tools, draft artifacts, execute bounded actions, and ask for human intervent…

This study analyzes recent actions by European regulators to support enterprise digitalization within the Single Market, focusing on small businesses. It highlights how Information and Communication Technologies (ICT), e…

The rapid integration of artificial intelligence (AI) into critical infrastructure including healthcare, finance, energy, and defense, offers transformative benefits but also conflicts with evolving regulatory and govern…

Artificial Intelligence deployment in enterprise environments has accelerated dramatically, yet governance frameworks struggle to keep pace with the technology's rapid evolution. This research develops a comprehensive ri…

This paper investigates the effectiveness of LLM critic models in improving agent reliability. It demonstrates that while LLM critics can have high offline accuracy, their interventions at deployment time can lead to sev…

AI agents often fail in ways that are difficult to localize due to probabilistic, long-horizon, multi-agent executions and noisy tool outputs. This paper addresses this by manually annotating failed agent runs and releas…

This paper frames LLM context window management as a constrained resource allocation problem. It proposes a declarative approach where sources contribute to the context proportionally, compressing large memory entries mo…

LLMs are increasingly used for Islamic question answering, where ungrounded responses may carry serious religious consequences. Yet standard MCQ/MRC-style evaluations do not capture key real-world failure modes, notably …

This paper explores conditional memory access to extend LLM context length while managing KV footprint. It evaluates long-context performance by increasing the Rotary Position Embedding base.…

This paper presents ARKV, an adaptive and resource-efficient KV cache management framework for LLM inference under limited memory budgets. It aims to reduce memory usage and maintain high throughput for large context win…

This paper discusses the application of agentic Artificial Intelligence (AI) systems to infrastructure-constrained environments, focusing on observability gaps, failure modes, and AI governance primitives. The study meas…

This paper describes AI agents as stochastic dynamical systems and frames learning to reason as transductive inference. It establishes three main results, including a possible failure mode of naive scaling where models b…

This paper empirically characterizes faults in agentic AI systems by analyzing 13,602 closed issues and merged pull requests from 40 open-source repositories. It derives taxonomies of fault types, symptoms, and root caus…

This paper presents the 2025 AI Agent Index, documenting information regarding the origins, design, capabilities, ecosystem, and safety features of 30 state-of-the-art AI agents. It illuminates broader trends in the deve…

We present an initial investigation into Agentic Retrieval-Augmented Generation (RAG) for Ukrainian, conducted within the UNLP 2026 Shared Task on Multi-Domain Document Understanding. Our system combines two-stage retrie…

This paper examines the legal implications of AI agents under existing and proposed EU law, particularly in light of the AI Act. It analyzes how current regulations apply to autonomous AI systems and identifies potential…

Artificial Intelligence (AI) deployment in healthcare is accelerating, yet governance frameworks remain fragmented and often assume extensive resources. Through a systematic review of 35 frameworks for AI implementation …

AI agents are increasingly deployed to execute important tasks, but many still fail in practice despite rising accuracy scores. This paper proposes twelve concrete metrics that decompose agent reliability along four key …

…

…

…

We present the PASF/PADE framework for agentic process automation in enterprise environments. The framework provides a systematic approach to identifying, designing, and deploying AI agents for long-horizon enterprise wo…

We present a comprehensive survey of enterprise AI agent adoption in 2025, covering deployment patterns, failure rates, governance practices, and the roadmap to Zone III autonomous operations. Based on interviews with 15…

We present the Agentification Factory methodology for systematically developing enterprise AI agents at scale. The factory model provides repeatable processes for agent design, testing, deployment, and governance.…

We present Patternomics, a framework for identifying, cataloguing, and reusing successful patterns in enterprise AI deployments. Patternomics provides a systematic approach to building institutional knowledge about what …

We present the Roundtrip Value Governance (RVG) framework, which ensures that every agentic workflow creates measurable value that can be traced back to the original business objective. RVG provides a closed-loop account…

We present a method for improving tool-calling agents at inference time through a reviewer agent that evaluates tool calls before execution. The reviewer provides feedback that allows the primary agent to correct its too…

We present a framework for understanding and optimizing the economic model of enterprise AI deployments, covering token costs, compute allocation, value attribution, and ROI measurement for agentic workflows.…

We present the Evidence Factory, a systematic approach to building and maintaining the research evidence base for enterprise AI decisions. The Evidence Factory automates research collection, synthesis, and relevance asse…

Modern code generation has made significant strides in functional correctness and execution efficiency. However, these systems often overlook a critical dimension in real-world software development: maintainability. To h…

This paper investigates the transformative potential of agentic AI in cybersecurity, specifically addressing how it can enhance practices in response to emerging threats. It aims to explore how agentic AI can transform c…

Large Language Models (LLMs) have rapidly gained popularity, transforming research and industry. To support their adoption, LLM agent workflow orchestration frameworks (hereinafter referred to as LLM agent frameworks) li…

The integration of Large Language Models (LLMs) into multi-agent systems has opened new avenues for automating complex workflows. However, designing robust software architectures for these systems remains a challenge. In…

Recent multi-agent frameworks built upon large language models (LLMs) have demonstrated remarkable capabilities in complex task planning. However, in real-world enterprise environments, business workflows are typically c…

Large Language Model (LLM)-based agentic systems have shown strong capabilities across various tasks. However, existing multi-agent frameworks often rely on static or task-level workflows, which either over-process simpl…

Traditional industrial automation systems require specialized expertise to operate and complex reprogramming to adapt to new processes. Large language models offer the intelligence to make them more flexible and easier t…

The integration of multi-agent systems (MAS) in software engineering is revolutionizing the way developers interact with artificial intelligence, shifting from simple code generation to sophisticated AI-driven collaborat…

In recent years, Large Language Models (LLMs) have achieved remarkable success and have been widely used in various downstream tasks, especially in the tasks of the software engineering (SE) field. We find that many stud…

Large Language Models (LLMs) have made significant strides in code generation and problem solving. Current approaches employ external tool-based iterative debuggers that use compiler or other tool-based runtime feedback …

The integration of Large Language Models (LLMs) into software engineering has driven a transition from traditional rule-based systems to autonomous agentic systems capable of solving complex problems. However, systematic…

Automated programming has become a powerful tool for solving real-world problems. Code generation, in particular, plays a key role in improving developer productivity and reducing the entry barrier to software developmen…

Large Language Models (LLMs) have shown impressive capabilities in generating human-like text, but they often suffer from "hallucinations," producing factually incorrect or nonsensical information. This issue severely li…

Large Language Models (LLMs) are increasingly deployed in real-world applications, yet their propensity for generating factually incorrect or inconsistent information, known as hallucination, remains a significant concer…

Large language models (LLMs) have transformed natural language processing, achieving remarkable performance across diverse tasks. However, their impressive fluency often comes at the cost of producing false or fabricated…

High-quality code documentation is crucial for software development especially in the era of AI. However, generating it automatically using Large Language Models (LLMs) remains challenging, as existing approaches often p…

Multi-agent Large Language Model (LLM) systems have been leading the way in applied LLM research across a number of fields. One notable area is software development, where researchers have advanced the automation of code…

Large language models (LLMs) struggle with complex, multi-step, long-horizon tasks. This paper proposes Plan-and-Act, a framework that incorporates explicit planning into LLM-based agents and enhances plan generation thr…

Existing benchmarks for LLM agents often focus on atomic tasks, failing to capture the complexities of real-world, long-horizon workflows. This paper introduces OdysseyBench, a comprehensive benchmark for evaluating LLM …

This paper proposes an end-to-end Agent Assessment Framework for evaluating agentic AI systems beyond traditional task completion metrics. It addresses the challenges posed by the non-deterministic nature of LLM agents a…

This survey provides a comprehensive overview of the evaluation methodologies for LLM-based agents. It categorizes existing approaches, discusses common challenges, and highlights key metrics used to assess agent perform…

Agents have experienced significant growth in recent years, largely due to the rapid technological advancements of Large Language Models (LLMs). Although these agents benefit from LLMs’ advanced generation proficiency, t…

Maintaining comprehensive and up-to-date knowledge graphs (KGs) is critical for modern AI systems, but manual curation struggles to scale with the rapid growth of scientific literature. This paper presents KARMA, a novel…

The ReAct capability in LLMs forms the foundation of modern agentic systems, but smaller models struggle with complex reasoning tasks. This paper introduces Pre-Act, a novel approach that enhances agent performance by cr…

LLM-based multi-agent systems often incur high latency for complex tasks requiring multiple iterative reasoning cycles. This paper proposes M1-Parallel, a framework that concurrently runs multiple multi-agent teams to un…

This study proposes a method for generating complex and long-horizon off-line task plans using large language models (LLMs). It addresses the limitation of simple planning results by enabling the LLM to actively collect …

AI agents have experienced a paradigm shift, from early dominance by reinforcement learning (RL) to the rise of agents powered by large language models (LLMs), and now further advancing towards a synergistic fusion of RL…

Autonomous agents have recently achieved remarkable progress across diverse domains, yet most evaluations focus on short-horizon, fully observable tasks. In contrast, many critical real-world tasks, such as large-scale s…

Large Language Model (LLM)-based agents have recently shown impressive capabilities in complex reasoning and tool use via multi-step interactions with their environments. While these agents have the potential to tackle c…

This paper explores the application of agentic generative AI for optimizing historical cost data in the construction industry. It focuses on developing context-aware outlier removal techniques to improve the accuracy of …

Recent advancements in large language models (LLMs) have significantly advanced the automation of software development tasks, including code synthesis, program repair, and test generation. More recently, researchers and …

This paper introduces a structured methodology for designing, developing, and deploying production-grade agentic AI workflows. It addresses challenges related to reliability, observability, maintainability, and alignment…

Integrating Large Language Models (LLMs) into autonomous agents marks a significant shift in the research landscape by offering cognitive abilities that are competitive with human planning and reasoning. This article exp…

Large language models (LLMs) are often praised for exhibiting near-human performance on a wide range of tasks. However, the rise of agentic AI systems is ushering in applications where language models perform specialized…

This blog post discusses the critical need for robust AI governance frameworks as agentic AI rapidly advances. It emphasizes that governance should go beyond mere compliance to define how organizations design, deploy, an…

The emergence of agentic AI systems, autonomous entities capable of reasoning, acting, and collaborating marks a significant shift in the evolution of artificial intelligence. These systems promise transformative benefit…

With the rise of generative AI, industry interest in software agents is growing. Given the stochastic nature of generative AI-based agents, their effective and safe deployment in organizations requires robust governance,…

Pre-trained on massive amounts of code and text data, large language models (LLMs) have demonstrated remarkable achievements in performing code generation tasks. With additional execution-based feedback, these models can…

The 2025 Artificial Intelligence and Business Strategy report, from MIT Sloan Management Review and Boston Consulting Group, looks at how organizations that are adopting agentic AI are gaining advantage while facing four…

This paper critically examines how the discourse around 'AI safety' is sometimes used to resist or undermine effective regulatory oversight of AI systems. It argues that an overemphasis on speculative, long-term risks ca…

This paper proposes a novel framework for 'superalignment' that extends beyond traditional weak-to-strong alignment by incorporating human-AI collaboration for robust oversight. It argues that as AI systems become more c…

This paper presents a comprehensive analysis of Agentic Gen Artificial Intelligence (AI) frameworks and their integration into modern military systems. We examine the architectural foundations, development pipelines, and…

The OWASP GenAI Security Project released the OWASP Top 10 for Agentic Applications, a key resource to help organizations identify and mitigate the unique risks posed by autonomous AI agents. This framework, developed wi…

Code generation agents powered by large language models (LLMs) are revolutionizing the software development paradigm. Distinct from previous code generation techniques, code generation agents are characterized by three c…

Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) by grounding responses with retrieved information. As an emerging paradigm, Agentic RAG further enhances this process by introducing autonomous L…

Large Language Models (LLMs) have demonstrated remarkable prowess in generating contextually coherent responses, yet their fixed context windows pose fundamental challenges for maintaining consistency over prolonged mult…

The concept of AI-native memory is revolutionizing how AI agents interact and learn, moving beyond stateless interactions to create persistent, context-aware 'Second Me' agents. This article explores the architectural sh…

The rise of LLM-based agents has opened new frontiers in AI applications, yet evaluating these agents remains a complex and underdeveloped area. This survey provides an in-depth overview of the emerging field of LLM agen…

Agentic AI systems powered by large language models (LLMs) and endowed with planning, tool use, memory, and autonomy, are emerging as powerful, flexible platforms for automation. Their ability to autonomously execute tas…

This study aims to analyze AI Risk Management Frameworks (AI RMFs), exploring their role in promoting the safe, accountable, and transparent adoption of AI technologies within economic systems. The first part of the rese…

This article challenges conventional boundaries between human and artificial cognition by examining introspective capabilities in large language models (LLMs). Although humans have traditionally been considered unique in…

With the widespread usage and application of Artificial Intelligence, the need to ensure responsible use of AI systems has become a subject of debate. This paper examines the comprehensive approach adopted by the EU AI A…

The year 2024 witnessed accelerated global AI governance advancements, marked by strengthened multilateral frameworks and proliferating national regulatory initiatives. This acceleration underscores an unprecedented need…

We present a novel pipeline, ReflectEvo, to demonstrate that small language models (SLMs) can enhance meta introspection through reflection learning. This process iteratively generates self-reflection for self-training, …

This paper proposes a novel agentic memory system for LLM agents that dynamically organizes memories using the Zettelkasten method. It creates interconnected knowledge networks through dynamic indexing and linking, allow…

Advanced large language models (LLMs) frequently reflect in reasoning chain-of-thoughts (CoTs), where they self-verify the correctness of current solutions and explore alternatives. However, given recent findings that LL…

Large Language Models (LLMs) have shown impressive capabilities in various natural language processing tasks, but complex multi-step mathematical reasoning remains a significant challenge. This paper introduces a novel a…

The emergence of AI agents powered by large language models (LLMs) marks a pivotal shift toward the Agentic Web, a new phase of the internet defined by autonomous, goal-driven interactions. In this paradigm, agents inter…

AI agents interacting with each other directly and across internet platforms and physical environments create security challenges beyond traditional cybersecurity and AI safety frameworks. This paper introduces multi-age…

Agentic AI and Large Language Models (LLMs) are transforming how language is understood and generated while reshaping decision-making, automation, and research practices. LLMs provide underlying reasoning capabilities, a…

Crafting response letters to reviewers’ comments is a time-consuming yet critical part of academic peer review. The inexperience of researchers can hinder the publication of their work, exacerbating the Matthew effect in…

This paper introduces Low Rank Query and Key attention (LRQK), a two-stage framework that decomposes full-precision query and key matrices into compact rank factors to reduce GPU memory costs for long-context inference. …

This chapter explores the cognitive reasoning processes involved in translation, comparing human capabilities with those of large language models (LLMs). We evaluate the effectiveness of various reasoning paradigms, incl…

This report serves as an accessible guide to the emerging field of AI agent governance. Agents - AI systems that can autonomously achieve goals in the world, with little to no explicit human instruction about how to do s…

This paper presents insights from a large-scale public competition designed to evaluate the security challenges in AI agent deployment. The competition involved 1.8 million prompt injection attacks and revealed 60,000 su…

Testing filter…

Testing approval…

We characterize the phenomenon of agent drift — the gradual degradation of semantic coherence in long-running autonomous AI systems. We identify three primary drift mechanisms: context contamination, goal displacement, a…

LangGraph is a library for building stateful, multi-actor applications with LLMs, used to create agent and multi-agent workflows. It extends LangChain with the ability to coordinate multiple chains (or actors) across mul…

We present a framework for evaluating AI agents on long-horizon tasks that require sustained reasoning, planning, and execution over extended periods. We identify key failure modes and propose metrics for measuring auton…

We introduce OdysseyBench, a benchmark specifically designed to evaluate language agents on tasks requiring hundreds of sequential reasoning steps. OdysseyBench reveals non-linear performance degradation with task length…

We survey the landscape of long-horizon planning with LLMs, covering task decomposition, subgoal generation, plan verification, and execution monitoring. We identify key challenges and promising research directions.…

We present a systematic framework for analyzing failures in AI agent systems, covering failure mode identification, root cause analysis, and mitigation strategy development. The framework is validated on 500+ real agent …

We present a comprehensive framework for governing autonomous AI systems, covering technical safety measures, organizational governance, and regulatory compliance. The framework addresses the unique challenges of long-ru…

A practitioner's guide to deploying agentic AI in enterprise environments, covering common patterns, failure modes, and architectural best practices from real-world deployments.…

We study human-AI collaboration patterns in long-horizon enterprise tasks, identifying optimal handoff points, oversight mechanisms, and collaboration protocols for Zone III workflows.…

We introduce the Model Context Protocol (MCP), an open standard that enables seamless integration between AI assistants and external data sources, tools, and services. MCP provides a universal protocol for AI-tool commun…

We examine the safety and governance requirements for deploying autonomous AI agents in healthcare settings, covering clinical decision support, administrative automation, and patient-facing applications.…

The Financial Stability Board examines the use of agentic AI in financial services, identifying opportunities for efficiency gains and systemic risks from autonomous agent deployment.…

RAISE implements a dual-component memory system enabling LLM agents to backtrack to previous states and handle interrupts, improving reliability in long-horizon tasks.…

Gartner examines the enterprise readiness for autonomous AI deployment, identifying the key technical, organizational, and governance challenges that must be addressed before Zone III workflows can be safely deployed at …

We present LAGRANGE, a framework for grounding agent actions in structured knowledge graphs to improve reliability and interpretability of long-horizon agent navigation tasks.…

We demonstrate that LLM agents can autonomously exploit one-day vulnerabilities in real systems, raising significant security concerns for autonomous AI deployment.…

Andrew Ng describes four key agentic design patterns — reflection, tool use, planning, and multi-agent collaboration — and argues that agentic workflows represent the next major wave of AI capability improvement.…

We survey methods for extending the context length of language models, covering positional encoding extensions, efficient attention mechanisms, and memory-augmented architectures.…

We prove that hallucination is an innate limitation of LLMs, not a fixable bug. We show that any LLM that can answer all questions must hallucinate on some inputs, establishing fundamental limits on factual reliability.…

Microsoft's comprehensive framework for responsible AI deployment in enterprise settings, covering fairness, reliability, privacy, security, inclusiveness, transparency, and accountability.…

The EU AI Act establishes a comprehensive regulatory framework for artificial intelligence in the European Union, with specific requirements for high-risk AI systems including autonomous agents in enterprise settings.…

We introduce SWE-bench, a benchmark for evaluating LLMs on real-world software engineering tasks. SWE-bench contains 2,294 GitHub issues and pull requests from 12 popular Python repositories.…

We study the challenges of long-horizon task planning for embodied agents and propose methods for improving robustness to execution failures and environmental changes.…

We propose CodeAct, an agent design that uses executable Python code as the action space instead of structured JSON actions. CodeAct agents can dynamically create and execute code to interact with environments.…

We survey the application of formal verification methods to language models and AI agents. We identify key challenges and promising approaches for providing formal guarantees about agent behavior.…

We present Self-RAG, a framework that trains LLMs to retrieve, generate, and critique their own outputs. Self-RAG adaptively retrieves passages and generates reflective tokens to improve output quality.…

We present a real-world case study of LLM serving infrastructure, analyzing latency, throughput, and reliability challenges in production deployments at scale.…

We survey the rapidly evolving landscape of AI agent frameworks, analyzing their architectural choices, capabilities, and limitations. We identify key trends and open challenges in agent framework design.…

Large language models (LLMs) have revolutionized natural language processing, but their tendency to "hallucinate"—generating factually incorrect or nonsensical information—remains a significant challenge. Current methods…

Hallucination has been widely recognized to be a significant drawback for large language models (LLMs). There have been many works that attempt to reduce the extent of hallucination. These efforts have mostly been empiri…

Business Process Management (BPM) aims to improve organizational activities and their outcomes by managing the underlying processes. To achieve this, it is often necessary to consider information from various sources, in…

As Large Language Models (LLMs) continue to advance in their ability to write human-like text, a key challenge remains around their tendency to hallucinate generating content that appears factual but is ungrounded. This …

With the rise of large language models (LLMs), researchers are increasingly exploring their applications in various vertical domains, such as software engineering. LLMs have achieved remarkable success in areas including…

Businesses are shifting away, however, and realizing agentic AI systems of autonomous, goal-oriented, software agents that are able to perceive, reason, make actions, and learn end-to-end systems. In this paper, a concep…

This paper presents a comprehensive survey on human-AI collaboration with large foundation models (LFMs). It categorizes existing research into three main types: human-in-the-loop, AI-in-the-loop, and human-AI co-creatio…

The rapid advancement of foundation models (FMs) across language, image, audio, and video domains has shown remarkable capabilities in diverse tasks. However, the proliferation of FMs brings forth a critical challenge: t…

We introduce Chain-of-Abstraction (CoA), a method for efficient tool use that separates reasoning from tool execution. CoA generates abstract reasoning chains first, then fills in tool calls, reducing redundant API calls…

We propose Mixture-of-Agents (MoA), a methodology that leverages the collective strengths of multiple LLMs to improve overall performance. MoA uses multiple LLMs as proposers and aggregators in a layered architecture.…

We propose an automated evaluation framework for multi-turn dialogue systems that assesses coherence, consistency, and task completion across extended conversations.…

We study how to optimally scale test-time compute for LLMs. We find that the optimal allocation of test-time compute depends on the difficulty of the problem and the capabilities of the model.…

We study optimal allocation of test-time compute for LLMs, showing that scaling inference compute can be more efficient than scaling training compute for many tasks.…

We study whether LLMs can be trusted as evaluators, finding systematic biases including position bias, verbosity bias, and self-enhancement bias that affect evaluation reliability.…

We present a platform and benchmark for long-horizon vision-language navigation, requiring agents to navigate complex environments over extended time horizons with minimal guidance.…

We propose a model selection framework for multi-modal reasoning that dynamically selects the most appropriate model based on task characteristics, improving robustness and efficiency.…

We introduce Mixtral 8x7B, a Sparse Mixture of Experts language model. Mixtral uses a router to select 2 of 8 expert FFN layers per token, achieving strong performance with reduced inference cost.…

Multi-Agent Reinforcement Learning (MARL) approaches have emerged as popular solutions to address the general challenges of cooperation in multi-agent environments, where the success of achieving shared or individual goa…

This paper provides a comprehensive review of existing methodologies for evaluating human-AI collaboration and proposes a new methodological framework. It identifies key dimensions for assessment, including task performa…

Temporal's durable execution model ensures that workflows continue executing despite failures, providing automatic retry, state persistence, and exactly-once semantics for long-running business processes.…

Durable execution is a programming model that ensures workflows survive failures, restarts, and infrastructure changes. Temporal provides durable execution primitives for building reliable long-running workflows.…

We present MemGPT, a system that intelligently manages different memory tiers in order to effectively provide extended context within the LLM's limited context window. MemGPT knows when to push critical information to a …

MetaGPT encodes software development SOPs into multi-agent collaboration, enabling agents to produce structured outputs (PRDs, architecture designs, code) through role-based communication.…

We apply systems safety engineering principles to autonomous AI systems, arguing that AI reliability requires the same rigorous engineering discipline as safety-critical systems in aerospace and nuclear industries.…

AutoGen is a framework for building LLM applications through multi-agent conversations, enabling flexible agent interaction patterns and human-in-the-loop capabilities.…

We present AutoGen, a framework that enables the development of LLM applications using multiple agents that can converse with each other to solve tasks. AutoGen agents are customizable, conversable, and seamlessly allow …

We introduce GAIA, a benchmark for general AI assistants that tests real-world capabilities requiring multi-step reasoning, tool use, and information synthesis. GAIA questions require an average of 5.4 steps to solve.…

WebArena is a standalone, self-hostable web environment for building autonomous agents. It includes four websites mimicking real-world applications and a benchmark of 812 long-horizon tasks.…

We present OpenAgents, an open platform for deploying and evaluating language agents in real-world settings. OpenAgents includes three agents: a data analysis agent, a plugin agent, and a web agent.…