{
 "A": {
  "title": "Position: Agentic AI System Is a Foreseeable Pathway to AGI",
  "authors": [
   "Junwei Liao",
   "Shuai Li",
   "Muning Wen",
   "Jun Wang",
   "Weinan Zhang"
  ],
  "n_authors": 5,
  "date": "2026/09/29",
  "venue": "International Conference on Machine Learning",
  "type": "",
  "code": [],
  "abstract": "Is monolithic scaling the only path to AGI? This paper challenges the dogma that purely scaling a single model is sufficient to achieve Artificial General Intelligence. Instead, we identify Agentic AI as a necessary paradigm for mastering the complex, heterogeneous distribution of real-world tasks. Through rigorous theoretical derivations, we contrast the optimization constraints of monolithic learners against the efficiency of Agentic systems, progressing from simple routing mechanisms to general Directed Acyclic Graph (DAG) topologies. We demonstrate that Agentic AI achieves exponentially superior generalization and sample efficiency. Finally, we discuss the connection to Mixture-of-Experts, reinterpret the instability of current multi-agent frameworks, and call for greater research focus on Agentic AI.",
  "url": "https://proceedings.mlr.press/v306/liao26f.html",
  "kind": "position paper (ICML 2026, PMLR v306)",
  "authors_shown": [
   "Junwei Liao",
   "Shuai Li",
   "Muning Wen"
  ]
 },
 "B": {
  "title": "Position: Agentic Systems Should be General",
  "authors": [
   "Elron Bandel",
   "Asaf Yehudai",
   "Alexandre Lacoste",
   "Avijit Ghosh",
   "Graham Neubig",
   "Margaret Mitchell",
   "Michal Shmueli-Scheuer",
   "Leshem Choshen"
  ],
  "n_authors": 8,
  "date": "2026/09/29",
  "venue": "International Conference on Machine Learning",
  "type": "",
  "code": [],
  "abstract": "We call for the development of agentic systems that thrive in new environments. Agentic systems, comprising foundation models, tools, and an execution strategy, have demonstrated strong capabilities, yet their development is often constrained by narrow benchmarks and their operation is siloed to limited environments. This paper advocates for developing general, adaptive agents that excel across diverse environments, from terminals and web interfaces to biological and embodied settings. We examine current limitations, explain the potential of increased generality, and identify immediate development priorities. Finally, we argue that protocols and evaluation must prioritize adaptiveness to foster a shared ecosystem for general-purpose agentic systems.",
  "url": "https://proceedings.mlr.press/v306/bandel26a.html",
  "kind": "position paper (ICML 2026, PMLR v306)",
  "authors_shown": [
   "Elron Bandel",
   "Asaf Yehudai",
   "Alexandre Lacoste"
  ]
 },
 "C": {
  "title": "Continual Learning Bench: Evaluating Frontier AI Systems in Real-World Stateful Environments",
  "authors": [
   "Asawa, Parth",
   "Glaze, Christopher M.",
   "Orlanski, Gabriel",
   "Ramakrishnan, Ramya",
   "Xu, Benji",
   "Biswal, Asim",
   "Chen, Vincent Sunn",
   "Sala, Frederic",
   "Zaharia, Matei",
   "Gonzalez, Joseph E."
  ],
  "n_authors": 10,
  "date": "2026/06/04",
  "venue": "2606.05661",
  "type": "",
  "code": [],
  "abstract": "Abstract: Continual learning, the ability of AI systems to improve through sequential experience, has attracted substantial interest, but no high-quality benchmark exists to evaluate it. We introduce Continual Learning Bench (CL-Bench), the first difficult, expert-validated benchmark designed to measure whether LLM-based systems genuinely improve with experience. CL-Bench spans six diverse domains (software engineering, signal processing, disease outbreak forecasting, database querying, strategic game-playing, and demand forecasting), each validated by domain experts and designed so that tasks share a learnable latent structure (codebase layout, disease outbreak dynamics, opponent strategies) that a stateful system can discover online but a stateless one cannot. We evaluate frontier models across several agent architectures, from naive in-context learning (ICL) to dedicated memory systems, introducing a gain metric to isolate learning from prior capabilities. We find that these systems leave headroom for improved continual learning: agents frequently overfit to immediate observations or fail to reuse knowledge across instances, and dedicated memory systems do not fix this -- in fact, naive ICL outperforms systems dedicated to memory management. CL-Bench is the first benchmark to evaluate continual learning across diverse real-world domains with expert-validated tasks and isolate online learning from underlying model capability, showing a need for better continual learning systems.",
  "url": "https://arxiv.org/abs/2606.05661",
  "kind": "benchmark preprint (arXiv); code github.com/pgasawa/continual-learning-bench",
  "authors_shown": [
   "Asawa, Parth",
   "Glaze, Christopher M.",
   "Orlanski, Gabriel"
  ]
 },
 "D": {
  "title": "MemoryArena: Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks",
  "authors": [
   "He, Zexue",
   "Wang, Yu",
   "Zhi, Churan",
   "Hu, Yuanzhe",
   "Chen, Tzu-Ping",
   "Yin, Lang",
   "Chen, Ze",
   "Wu, Tong Arthur",
   "Ouyang, Siru",
   "Wang, Zihan",
   "Pei, Jiaxin",
   "McAuley, Julian"
  ],
  "n_authors": 14,
  "date": "2026/02/18",
  "venue": "2602.16313",
  "type": "",
  "code": [],
  "abstract": "Abstract: Existing evaluations of agents with memory typically assess memorization and action in isolation. One class of benchmarks evaluates memorization by testing recall of past conversations or text but fails to capture how memory is used to guide future decisions. Another class focuses on agents acting in single-session tasks without the need for long-term memory. However, in realistic settings, memorization and action are tightly coupled: agents acquire memory while interacting with the environment, and subsequently rely on that memory to solve future tasks. To capture this setting, we introduce MemoryArena, a unified evaluation gym for benchmarking agent memory in multi-session Memory-Agent-Environment loops. The benchmark consists of human-crafted agentic tasks with explicitly interdependent subtasks, where agents must learn from earlier actions and feedback by distilling experiences into memory, and subsequently use that memory to guide later actions to solve the overall task. MemoryArena supports evaluation across web navigation, preference-constrained planning, progressive information search, and sequential formal reasoning, and reveals that agents with near-saturated performance on existing long-context memory benchmarks like LoCoMo perform poorly in our agentic setting, exposing a gap in current evaluations for agents with memory. MemoryArena is now released at this https URL .",
  "url": "https://arxiv.org/abs/2602.16313",
  "kind": "benchmark preprint (arXiv, comment: ICML 2026); code github.com/ZexueHe/MemoryArena",
  "authors_shown": [
   "He, Zexue",
   "Wang, Yu",
   "Zhi, Churan"
  ]
 },
 "E": {
  "title": "AMA-Bench: Evaluating Long-Horizon Memory for Agentic Applications",
  "authors": [
   "Yujie Zhao",
   "Boqin Yuan",
   "Junbo Huang",
   "Haocheng Yuan",
   "Zhongming Yu",
   "Haozhou Xu",
   "Lanxiang Hu",
   "Abhilash Shankarampeta",
   "Zimeng Huang",
   "Wentao Ni",
   "Yuandong Tian",
   "Jishen Zhao"
  ],
  "n_authors": 12,
  "date": "2026/09/29",
  "venue": "International Conference on Machine Learning",
  "type": "",
  "code": [
   "https://github.com/AMA-Bench/AMA-Bench"
  ],
  "abstract": "Large Language Models (LLMs) are increasingly used as autonomous agents in complex, long-horizon applications, where effective memory is critical for sustained performance. Yet existing memory benchmarks are largely dialogue-centric, while real agent memory consists of continuous agent-environment interaction trajectories composed of states, actions, observations, and tool outputs. To address this gap, we introduce AMA-Bench ( A gent M emory with A ny length), a benchmark for evaluating long-horizon memory in realistic agentic settings. AMA-Bench combines real-world agent trajectories from representative applications with expert-curated QA, as well as synthetic trajectories that scale to arbitrary horizons with rule-based QA. Our study shows that existing memory systems underperform because they fail to capture causal and objective information and rely heavily on lossy similarity-based retrieval. We further propose AMA-Agent , a memory system based on causality-graph construction and tool-augmented retrieval. AMA-Agent achieves 57.22% accuracy on AMA-Bench, outperforming the strongest baseline by 11.16% . Resources are available at: https://ama-bench.github.io/.",
  "url": "https://proceedings.mlr.press/v306/zhao26bs.html",
  "kind": "benchmark paper (ICML 2026, PMLR v306); code github.com/AMA-Bench/AMA-Bench",
  "authors_shown": [
   "Yujie Zhao",
   "Boqin Yuan",
   "Junbo Huang"
  ]
 },
 "F": {
  "title": "AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs",
  "authors": [
   "Shu, Raphael",
   "Zhang, Yusen",
   "Cho, Young Min",
   "Yang, Jin Mo",
   "Yuan, Yuan",
   "Zheng, Wenliang",
   "Guntuku, Sharath Chandra",
   "Ungar, Lyle",
   "Yu, Zhou",
   "Zhang, Rui"
  ],
  "n_authors": 10,
  "date": "2026/09/25",
  "venue": "2609.31590",
  "type": "",
  "code": [],
  "abstract": "Abstract: Existing multi-agent benchmarks primarily test in competitive settings, short-horizon interactions under 20 steps, or simply aggregate individual performance, failing to isolate and highlight genuine collaboration capabilities of LLM-based agents. We introduce AgentWorld, a benchmark of 100 human-annotated tasks (with 100 augmented variants) for evaluating long-horizon, multi-agent collaboration. Tasks span 50+ interaction rounds across a rich MMORPG sandbox and require 3-20 agents with asymmetric roles and abilities to coordinate through communication, joint planning, and resource sharing under a blackbox setting where each agent acts independently without access to others' internal states. To quantify collaboration effectiveness in addition to conventional binary task success, we propose Causal Collaboration Effectiveness (CCE), a graph-based metric that traces causal dependencies between agent actions and measures what fraction of a team's effort actually contributed to the outcome. Experiments with Gemini 3 Flash, Claude Haiku 4.5, GPT-5 Mini, and DeepSeek R1-70B show that even the best model achieves only 52.0% task success, with systematic failure modes including communication breakdowns, role confusion, and inability to maintain shared plans across rounds. AgentWorld is fully open-source.",
  "url": "https://arxiv.org/abs/2609.31590",
  "kind": "benchmark preprint (arXiv, accepted COLM 2026)",
  "authors_shown": [
   "Shu, Raphael",
   "Zhang, Yusen",
   "Cho, Young Min"
  ]
 },
 "G": {
  "title": "AgentLAB: Benchmarking LLM Agents against Long-Horizon Attacks",
  "authors": [
   "Tanqiu Jiang",
   "Yuhui Wang",
   "Jiacheng Liang",
   "Ting Wang"
  ],
  "n_authors": 4,
  "date": "2026/09/29",
  "venue": "International Conference on Machine Learning",
  "type": "",
  "code": [
   "https://github.com/TanqiuJiang/AgentLAB"
  ],
  "abstract": "LLM agents are increasingly deployed in long-horizon, complex environments to solve challenging problems, but this expansion exposes them to long-horizon attacks that exploit multi-turn user\u2013agent\u2013environment interactions to achieve objectives infeasible in single-turn settings. To measure agent vulnerabilities to such risks, we present AgentLAB, the first benchmark dedicated to evaluating LLM agent susceptibility to adaptive, long-horizon attacks. Currently, AgentLAB supports five novel attack types including intent hijacking, tool chaining, task injection, objective drifting, and memory poisoning, spanning 28 realistic agentic environments, and 644 security test cases. Leveraging AgentLAB, we evaluate representative LLM agents and find that they remain highly susceptible to long-horizon attacks; moreover, defenses designed for single-turn interactions fail to reliably mitigate long-horizon threats. We anticipate that AgentLAB will serve as a valuable benchmark for tracking progress on securing LLM agents in practical settings. The code for AgentLAB is available at: https://tanqiujiang.github.io/AgentLAB_main.",
  "url": "https://proceedings.mlr.press/v306/jiang26as.html",
  "kind": "benchmark paper (ICML 2026, PMLR v306); code github.com/TanqiuJiang/AgentLAB",
  "authors_shown": [
   "Tanqiu Jiang",
   "Yuhui Wang",
   "Jiacheng Liang"
  ]
 },
 "H": {
  "title": "An Architecture for Long-Horizon Agents: Levels, Ticks and Cascaded Intelligence",
  "authors": [
   "Nijkamp, Erik",
   "Koul, Anurag",
   "Pakhomov, Egor",
   "Pang, Bo"
  ],
  "n_authors": 4,
  "date": "2026/09/17",
  "venue": "2609.19519",
  "type": "",
  "code": [],
  "abstract": "Abstract: Language-model agents are increasingly asked to carry out work spanning days or weeks, such as an operations remediation or a research programme. Such a task outlives any context window, any process and any interval at which a person can attend. In this paper, we argue that a long-horizon agent must run continually without forgetting before it can learn continually. This ability lies in the harness around the model rather than in the model itself. We derive seven bottlenecks from the long-horizon setting and answer them with a hierarchical architecture of three parts: (i) levels indexed by time scale, each keeping a bounded file summarising the level below; (ii) a clocked tick as the unit of autonomous action; and (iii) cascaded intelligence, where work is escalated to a more capable model only after failing review. We report on a ten-day campaign in which an agent built on this architecture reproduced a published reinforcement-learning result with a human attending once a day, and show (1) the agent kept the thread across every context reset and session boundary of the campaign, (2) operating knowledge written early changed later behaviour with no change to model weights, and (3) where learned components would enter such a system. Overall, our experience suggests continual learning for these agents needs a substrate outliving every context and process, and the checks the harness already runs are where a learner belongs.",
  "url": "https://arxiv.org/abs/2609.19519",
  "kind": "architecture preprint (arXiv) with a ten-day case study",
  "authors_shown": [
   "Nijkamp, Erik",
   "Koul, Anurag",
   "Pakhomov, Egor"
  ]
 },
 "I": {
  "title": "Accelerating scientific discovery with Co-Scientist",
  "authors": [
   "Gottweis, Juraj",
   "Weng, Wei-Hung",
   "Daryin, Alexander",
   "Tu, Tao",
   "Sirkovic, Petar",
   "Myaskovsky, Artiom",
   "Glowaty, Grzegorz",
   "Weissenberger, Felix",
   "Orlandi, Alessio",
   "Popovici, Dan",
   "Palepu, Anil",
   "Rong, Keran"
  ],
  "n_authors": 51,
  "date": "2026/07",
  "venue": "Nature",
  "type": "Article",
  "code": [],
  "abstract": "Scientific discovery is driven by scientists generating hypotheses for complex problems that undergo rigorous experimental validation. To augment this process, we introduce Co-Scientist, a multi-agent artificial intelligence (AI) system built on Gemini for structured scientific thinking and hypothesis generation. Co-Scientist aims to help scientists discover new original knowledge. Conditioned on their research objectives and previous scientific evidence, it formulates demonstrably novel research hypotheses for experimental verification. The system\u2019s design involves agents continuously generating, critiquing and refining hypotheses accelerated by scaling test-time compute. Key contributions include (1) a multi-agent architecture with an asynchronous task execution framework for flexible compute scaling, and (2) a tournament evolution process for self-improving hypotheses generation. Automated evaluations show continued benefits of test-time compute scaling, improving hypothesis quality over time. Although this is a general-purpose system, we focus the validation in three biomedical applications: drug repurposing; novel-target discovery 1 ; and explaining mechanisms of antimicrobial resistance 2 . Specifically, Co-Scientist helped to identify new drug-repurposing candidates and synergistic combination therapies for acute myeloid leukaemia that were validated through in vitro experiments. These real-world validations demonstrate the potential of Co-Scientist to accelerate scientific discovery and usher in an era of AI-empowered scientists.",
  "url": "https://doi.org/10.1038/s41586-026-10644-y",
  "kind": "peer-reviewed article (Nature)",
  "authors_shown": [
   "Gottweis, Juraj",
   "Weng, Wei-Hung",
   "Daryin, Alexander"
  ]
 },
 "J": {
  "title": "Agentic AI systems for scientific discovery",
  "authors": [
   "Feng, Bin",
   "Walther, Peter",
   "Zhang, Leili",
   "Coyle, Reid A.",
   "Kim, Seongmin",
   "Mandal, Abhijoy",
   "Zhang, Zijian",
   "Papidocha, Sven M.",
   "Kelmanson, Anna",
   "Segura, David Ming",
   "Thakur, Saumya",
   "Pal, Shyam Chand"
  ],
  "n_authors": 19,
  "date": "2026/10/22",
  "venue": "Nature Reviews Methods Primers",
  "type": "Primer",
  "code": [
   "https://github.com/K-Dense-AI/scientific-agent-skills"
  ],
  "abstract": "Artificial intelligence (AI) agents are emerging in scientific research as systems that can transform the accumulated decisions of scientists into traceable workflows. In fact, modern scientific practice depends on continual coordination of knowledge dispersed across literature, data, theory and experiment, which, however, is often made through manual, case-by-case judgement. This Primer introduces agentic AI systems as a general method to structure this process through natural language interfaces, serving as a computational brain in the laboratory. Central to this framework are practical workflows for in-depth literature search and knowledge mining, orchestration of physics-based simulations, and integration with automated experimental platforms. We begin by describing the foundations of AI agents, including task decomposition, memory, tool use and communication between agents, and by distinguishing single-agent and multi-agent designs, as well as fully autonomous and human-in-the-loop configurations. Equally important are the means by which these systems are evaluated, documented and reported reproducibly. Finally, we examine major bottlenecks, including hallucination, tool misuse, limited interpretability and safety risks, and outline a future in which coordinated networks of agents, instruments and researchers transform scientific discovery from isolated tasks into continuous adaptive experimentation.",
  "url": "https://www.nature.com/articles/s43586-026-00534-y",
  "kind": "primer / review (Nature Reviews Methods Primers)",
  "authors_shown": [
   "Feng, Bin",
   "Walther, Peter",
   "Zhang, Leili"
  ]
 },
 "K": {
  "title": "Autonomous Research Agents: A Survey of AI Scientists and the Verification Gap",
  "authors": [
   "Ding, Tianyu",
   "Nannapaneni, Aditya",
   "Liu, Bingfan",
   "Zhang, Ling"
  ],
  "n_authors": 4,
  "date": "2026/06/29",
  "venue": "2608.05179",
  "type": "",
  "code": [],
  "abstract": "Abstract: Large language model (LLM) agents are increasingly used across the scientific research lifecycle: ideation, literature search, experiment design and execution, analysis, manuscript drafting, and review. End-to-end AI scientist systems can now produce paper-like manuscripts, but their claims are often harder to verify than their code is to run. This survey studies that gap in computational AI/ML research, where code, benchmarks, experiments, and write-ups are most visible. We screen 125 candidate works and include 35, with full-text coding of 26 entries: 24 runnable systems and two study or position papers. We code seven audit dimensions: lifecycle stage, autonomy level, evaluation method, released artifacts, human-in-the-loop points, novelty verification, and result-selection disclosure. The main pattern is that code release is now common, but reproducibility-grade and claim-verification artifacts remain much less common. In the 24 runnable systems, 83 percent release code, while 38 percent release seeds or execution traces and 38 percent report any novelty-verification method. Among nine closed-loop L4 systems, seven are mechanical reruns and one is author-claimed without an external check; no LLM-era system in the corpus demonstrates an externally validated in-loop oracle under our coding rule. We contribute a coded corpus, a lifecycle-by-autonomy map, an auditability-gap analysis, and a reviewer-facing reporting checklist. The survey argues that the field's central bottleneck is no longer only whether agents can complete research tasks, but whether reviewers can verify the claims those agents produce.",
  "url": "https://arxiv.org/abs/2608.05179",
  "kind": "survey preprint (arXiv)",
  "authors_shown": [
   "Ding, Tianyu",
   "Nannapaneni, Aditya",
   "Liu, Bingfan"
  ]
 },
 "L": {
  "title": "Agent Memory: Characterization and System Implications of Stateful Long-Horizon Workloads",
  "authors": [
   "Omri, Yasmine",
   "Gan, Ziyu",
   "Broveak, Zachary",
   "Geens, Robin",
   "He, Zexue",
   "Pentland, Alex",
   "Verhelst, Marian",
   "Weissman, Tsachy",
   "Tambe, Thierry"
  ],
  "n_authors": 9,
  "date": "2026/06/04",
  "venue": "2606.06448",
  "type": "",
  "code": [],
  "abstract": "Abstract: LLM agents are increasingly deployed on long-horizon tasks requiring sustained reasoning over extended interaction histories. Realizing this at scale requires agents to persistently store, retrieve, and update their own memory across sessions. A rich ecosystem of agent memory systems has emerged spanning flat retrieval, LLM-mediated extraction, consolidating fact stores, and agentic control flows. Yet, their system-level behavior remains uncharacterized. We present the first systems characterization of agent memory. First, we introduce a system-oriented taxonomy classifying agent memory systems along four axes. Second, we build a phase-aware profiling harness attributing cost to construction, retrieval, and generation. Third, we characterize ten representative systems across two benchmark suites, uncovering how design choices shift cost across the write and read paths. Finally, we derive 10 system recommendations covering construction scheduling, capability floors, amortization via query volume, freshness-latency tradeoffs, and fleet-scale management.",
  "url": "https://arxiv.org/abs/2606.06448",
  "kind": "systems characterization preprint (arXiv)",
  "authors_shown": [
   "Omri, Yasmine",
   "Gan, Ziyu",
   "Broveak, Zachary"
  ]
 }
}