Agent 进展

Agentic AI development today runs on token maxing: buying capability with tokens — longer reasoning traces, more turns, wider tool payloads, bigger replayed contexts — so tokens per task grow faster than task value. Falling per-token prices mask the pattern; total spend rises anyway. We argue the decisive lever against token maxing is the harness: the orchestration layer that assembles context, exposes tools, sequences turns, delegates work, and carries enterprise observability and governance. We isolate it with a controlled swap: 22 locked evaluation tasks, six foundation models (Claude Sonnet 4.6, Gemini 3.1, Gemini Flash 3.5, Qwen 3.6, GLM 5.1, Palmyra X6), changing only the orchestration layer — a frozen conventional production loop versus the Writer Agent Harness. Holding models constant, the harness cuts blended cost per task 41% ($0.21->$0.12), median wall-clock 44% (48s->27s), and tokens per task 38% (14.2k->8.8k), with task-completion quality at parity (0.78->0.81, directional at this sample size). Efficiency is model-invariant — every model gets cheaper (33-61%) — while quality gains are capability-dependent: a model's gain correlates almost perfectly with its baseline strength (r=0.99, n=6), a phenomenon we term harness leverage. Quality per dollar rises 82%; task-completions per million tokens rise from 54.9 to 92.0. On this workload the orchestration layer moved cost per task more than the full spread of the model menu did. We formalize token economics at the orchestration layer (including effective input price under prompt caching), detail the six mechanism families behind the effect — cache-shape discipline to failure-spend governance — compare six widely used agent systems on the same axes, and argue the harness is the one component whose efficiency multiplies across every model an organization runs — present and future.

Large language models (LLMs) increasingly improve their reasoning at test time via additional computation, yet most existing works treat each problem in isolation. When problems arrive sequentially, accumulating reusable experience across them can further improve performance. Existing memory-based methods either store whole-solution templates that generalize poorly to novel problems or use heuristic step-level selection that is not optimized for final-answer correctness. Learning selection policies requires large-scale training data and fixed action spaces, making such approaches unsuitable for test-time settings where memory expands incrementally and only limited supervision is available. We propose MILES (Modular Instruction Memory with LEarnable Selection for self-improving LLM reasoning), a framework that dynamically expands step-wise memory and applies correctness-optimized memory composition under realistic test-time constraints. MILES maintains modular memory units consisting of asymmetric pairs of sub-goal embeddings and sub-instructions, each associated with a learnable selection head. This memory structure enables a coarse-to-fine retrieval mechanism: The coarse level enables memory expansion and collects supervision for training selection heads from confident samples, while the fine stage applies learned selection heads to rerank coarse-level candidates and guide reasoning for uncertain samples. MILES consistently matches or outperforms prior methods while achieving superior accuracy-efficiency tradeoffs. Extensive experiments demonstrate its effectiveness, robustness, and transferability.

Large language models are rapidly moving towards closing the development cycle, transitioning from simple assistive companions to autonomous contributors deeply embedded into collaborative development environments. Despite their accelerated adoption, existing evaluation techniques are limited due to their fragmented nature and distorted projection of true model capabilities, often obtained from hypothetical syntactic scenarios. This research aims to bridge this gap by providing a comprehensive evaluation methodology for LLM-powered agents that is grounded in real-world software development practice. Our evaluation approach focuses on contamination-awareness, in-the-wild agentic behavior assessment, and trajectory-aware benchmarks and metrics capturing realistic coding contexts, human-aligned behavior, and model failure modes.

Recent progress on ARC-AGI-1 from disclosed architectures has come broadly from two regimes: heavy test-time compute over frontier models (evolutionary search, exhaustive sampling, extended chain-of-thought), or benchmark-specific training in which small models are fine-tuned on ARC data, often with task-specialized architectures. We study a third regime: an open-weight model in non-thinking mode (DeepSeek V3.2) under a strict budget, with no ARC-specific fine-tuning. We study what is recoverable through architecture alone, building agentic harnesses that decompose pattern-discovery and program-synthesis stages explicitly. First, we introduce an Explorer-Definer Pipeline that separates pattern discovery from executable transformation synthesis, implemented as a two-stage agent pipeline. Next, we present the Reflective Orchestrator, which augments the pipeline with autonomous exploration of new transformations when previous hypotheses fail on training pairs. On the ARC-AGI-1 public 400-task evaluation set, the pipeline reaches 57.50% pass@2 at \$0.25 per task, and the orchestrator reaches 67.25% pass@2 at \$0.62 per task. Together these architectures lift a 15.50% one-shot baseline by ~52 points without benchmark-specific training or heavy test-time compute. Furthermore, the orchestrator-driven lift tests a falsifiable diagnostic the pipeline produces; unbiased pass@k analysis suggests the pipeline is generation-bound, not selection-bound (selection via training-pair accuracy captures ~95% of the candidate ceiling) and predicts that significant improvement requires broader generation, not better ranking. The orchestrator implements this prediction via adaptive re-exploration and confirms it (unbiased pass@1 lift +9.81 pp, matching selection-mediated pass@2 lift). An additional pipeline ablation identifies its think tool as a significant component, with removal reducing pass@2 by 5.75 pp.

We present AgentLens, a production-assessed benchmark for interactive code agents. Most code-agent benchmarks reduce a run to a single bit — did the task pass? — but the people who actually use these agents experience the entire trajectory: how the agent follows instructions, uses its tools, verifies its own work, recovers from mistakes, and talks to them along the way. AgentLens evaluates that whole trajectory. It pairs formal verification, where an objective check exists, with LLM-written trajectory reviews and side-by-side comparisons, so that each run yields a readable explanation of why the score is what it is. This makes AgentLens useful for more than ranking models: we use it to diagnose model behavior, compare successive versions of our own agent, and catch product regressions in a nightly evaluation pipeline. We release the benchmark as open source at https://github.com/agent-lens/agent-lens-bench.

Recent advances in AI for Mathematics have focused largely on autoformalization and theorem proving, leaving the role of Computer Algebra Systems (CAS) in agentic LLM workflows underexplored. We propose a ReAct-style agentic setup that combines LLM reasoning with verifiable feedback from SageMath, together with Context7 for the up-to-date documentation. We evaluate this agentic setup across frontier models for solving research-level mathematical problems from the RealMath benchmark in a setting that emulates a computational-mathematics research loop. We also propose a refinement to the RealMath benchmark by introducing a multi-step post-processing procedure and a multi-stage validation pipeline, both of which improve the quality and reliability of the extracted problem set. Our experiments reveal substantial performance gains from SageMath access across all evaluated models on +9.7~pp on average, the gains range from 1.5~pp to 27.8~pp and narrow the gap between open-weight and closed models. Qwen~3.7-Max benefits from SageMath the most, while GPT-5.5 achieves the highest solve rate of $75.2%$ and the lowest token usage among tool-enabled configurations. Our findings suggest that CAS-augmented agents represent a promising direction for assisting mathematicians in computational exploration, and we believe that this work is a step towards automated conjecture discovery. The project repository is available online.

Scientific results produced by LLM generated analysis code must be understandable and reproducible. However, uncertainty can arise at different stages of the process, both in the original natural language specification and in the generated implementation. As a result, even executable code may not provide a clear understanding of which quantities are being computed or which assumptions determine the final results. To address this challenge, we introduce quantity grounded semantic differencing, a multi-agent framework for analyzing and comparing scientific programs generated by LLMs. The framework assigns code generation, execution, tracing, and validation to separate agents, allowing it to reconstruct how key output quantities are produced and to identify differences between the intended analysis and the implemented code. We also introduce a module that inspects ambiguities in the initial user instruction and suggests alternative rewrites before code generation. Its modular design enables application to different scientific domains by replacing domain specific resources while preserving the same workflow. We validate the framework on representative collider physics analyses. The results demonstrate that the modular task decomposition enhances both transparency and reliability relative to the previous single prompt approach, while enabling substantially smaller models to execute the complete workflow.

Large language model coding agents increasingly perform open-ended data modeling and analysis. These agents are stochastic and adaptive, and therefore their autonomous model discovery behavior cannot be adequately characterized by a single benchmark run. In this work, we propose an experimental design and analysis framework for systematically evaluating this discovery process, quantifying its variability, and identifying important factors. The proposed framework treats these agents as stochastic model-discovery operators, which map task-specific discovery data and an optimization target to a fitted model. Specifically, we investigate two such operators, Codex and Claude Code, under controlled experimental factors including agent's reasoning effort, task, optimization metric, and composition of training data. For each agent-task-metric combination, regression models and inference are conducted for multiple responses such as output quality, dollar cost, wall-clock time, and process complexity. Furthermore, we develop a utility-aligned canonical decomposition to characterize the dominant direction of the reasoning-effort effect and to assess whether that direction aligns with a performance-cost utility direction. The proposed framework is demonstrated on a testbed of networked word-forming games with insightful findings on reasoning effort with respect to cost and process complexity.

Reinforcement learning has become a promising paradigm for improving large language model (LLM) agents on long-horizon search tasks, where the agent must make a sequence of intermediate decisions before receiving a final outcome. However, existing methods still face a key limitation: the rollout budget is often allocated without explicitly assessing the utility of intermediate states. As a result, substantial computation may be spent on low-value states, even though different branches can vary drastically in their informativeness. In this paper, we propose Information Gain-based Rollout Policy Optimization (IGRPO), a policy optimization framework that treats intermediate-state informativeness as the organizing principle of rollout collection. Specifically, IGRPO performs budget-aware tree-structured rollouts by allocating expansion budget according to node-level informativeness, so that more informative branches are expanded more frequently while unpromising branches are progressively suppressed. We further demonstrate that the information gain-based rollout induces an explicit limiting teacher distribution over trajectories, which naturally yields a clear policy optimization target, thereby unifying adaptive tree-structured exploration with principled policy learning under a single framework. Experiments on seven challenging search-augmented QA benchmarks demonstrate that IGRPO consistently outperforms strong baselines under the same rollout budget constraints, validating the effectiveness of leveraging the induced teacher distribution to guide policy optimization for long-horizon search agents.

Large language model (LLM) agents solving multi-step tasks frequently commit to trajectories that are doomed to fail, yet continue to consume substantial inference compute before the failure becomes observable. We show that failure is predictable early from the agent's internal representations: lightweight per-round probes on hidden activations anticipate eventual episode failure as early as the first interaction round, where scorers reading only the agent's observable behavior are barely better than chance. We turn this signal into a practical abort cascade: one distribution-free calibrated gate per round, with per-round recall budgets jointly searched so that eventually-successful episodes survive all gates at a user-specified global rate; this episode-level guarantee is the one that matters in deployment, since false-abort risk accumulates across gates. Across two agent models on TextCraft, the cascade meets every recall target from 90% to 97% and, at the 90% target, saves 47.1% +/- 10.3% (Qwen-2.5-7B) and 37.2% +/- 8.8% (Llama-3.2-3B) of inference compute, 1.6--1.7x the best single-gate policy. An otherwise-identical cascade reading only behavior saves roughly half as much, and adding behavioral features to the probe yields no further gain: the hidden states capture what behavior reveals. Finally, we characterize the sample complexity of certifying high recall targets, telling practitioners which recall promises their data can, and provably cannot, back. The code will be released soon.

Tool-augmented large language models extend their capabilities beyond parametric knowledge through external tools, but tend to invoke them unnecessarily. We investigate whether tool-use decisions have any stable internal representation that can be extracted and manipulated, a question that is non-trivial given that tools exist entirely in context at inference time and have no direct encoding in model weights. We show that steering vectors extracted from heading-anchors positions exert bidirectional causal control over tool-invocation behavior across five open-source models and three domains, suppressing unnecessary tool use most effectively in domains where parametric reasoning suffices. However, geometric analysis reveals that this causal effectiveness does not correspond to clean linear structure: tool-invocation steps exhibit diffuse, bimodal alignment with the suppression vector rather than the consistent negative alignment a linear encoding account would predict, and different tool types recruit largely distinct internal signatures with low cross-tool feature overlap. We hypothesize these geometric properties are indicative of the non-parametric nature of tools, and distinguish tool-use steering vectors from those extracted for parametrically grounded concepts. The relationship between this geometric irregularity and the observed causal effectiveness remains an open question.

Long-term user memory is essential for personalized conversational agents, yet many memory systems still expose memory through passive retrieval interfaces, making the model a consumer of pre-selected evidence. We introduce NapMem, a framework for learning to use long-term user memory as a structured action space rather than passively retrieved context. NapMem organizes user history into a linked multi-granularity memory pyramid, where raw conversations, typed memory records, topic tracks, and user profiles are connected through provenance relations, and exposes these levels through memory tools. The agent is trained to select memory according to the query and intermediate evidence, allowing it to inspect different memory granularities before answering. Experiments on PersonaMem-v2, LongMemEval, and LoCoMo show that a NapMem agent trained with memory-tool reinforcement learning is competitive across diverse memory-intensive tasks, while evaluations on non-memory tasks suggest that the learned policy largely preserves general reasoning and tool-use abilities. Additional analyses examine storage, inference cost, tool-use behavior, and ablations over navigation, memory granularity, and RL training. Our results suggest that long-term user memory benefits from coupling structured storage with a learned policy for using memory at the appropriate granularity.

Detecting vulnerability-inducing commits (VICs) at submission time is critical for improving the security and reliability of software systems. However, this task is highly challenging because it requires reasoning about the semantic impact of code changes from heterogeneous information sources, including code diffs, commit messages, and the surrounding contextual code. Existing approaches often struggle to fully capture these complex interactions, resulting in limited detection performance. In this paper, we propose VIC-RAGENT, an LLM-based multi-agent framework for effective and explainable vulnerability detection. VIC-RAGENT leverages multiple specialized agents to provide complementary perspectives, including structural analysis, intent understanding, and vulnerability inspection. To further improve detection reliability, the framework employs a multi-stage reasoning process that progressively refines candidate vulnerabilities through preliminary inspection, reanalysis, and a final decision stage. Experimental results on a real-world dataset across multiple LLMs demonstrate that VIC-RAGENT consistently outperforms baselines, including Direct, CoT, and CodeAgent. Compared to the strongest baseline, VIC-RAGENT achieves 1.2-1.7x higher F1-scores across different models. Overall, VIC-RAGENT offers a robust, explainable, and practical solution for detecting VICs in modern software development workflows.

Large language model (LLM) agents require post-training methods that can improve long-horizon decision making from environment feedback. However, existing agentic post-training pipelines often treat data curation as a fixed preprocessing step, focusing mainly on data augmentation while neglecting filtering, refinement, and adaptation to downstream failures. We propose CurateEvo, a failure-driven dynamic evolution framework for agentic post-training data curation. CurateEvo represents the curation strategy as executable code and iteratively rewrites it using failed trajectories from a held-out development set. At each epoch, the evolved strategy transforms a fixed raw corpus into supervised fine-tuning data, reinforcement learning data, and an inference-time memory bank. The evolution process first improves effectiveness by diagnosing recurring failure modes and augmenting, filtering, or refining data accordingly, and then improves efficiency by pruning redundant or low-utility training turns under a cost-aware objective. Experiments on ACEBench-Agent, BFCL-V4, and τ^2-Bench under both labeled and wild-data settings show that CurateEvo consistently outperforms prior curation methods, improving average scores by 3.2 and 2.7 points, respectively. Further analyses demonstrate that CurateEvo is compatible with different post-training recipes and substantially reduces curation overhead.

Formal verification offers the strongest guarantee of software correctness, but it does not scale: the proofs demanded by interactive theorem provers such as Coq require enormous expert effort. Large language models (LLMs) promise to generate these proofs automatically, yet existing approaches wire a fixed, human-designed proof strategy into the system and constrain the model to follow it (retrieving premises and predicting tactics one step at a time, or splitting goals by divide-and-conquer), and still prove only a fraction of their target theorems. We show that imposing such a strategy is unnecessary and limiting. Handing the whole lemma to a general LLM code agent (for example, Claude Code), free to choose its own approach, and wrapping it in a verification harness is both simpler and more effective, achieving full coverage: every targeted lemma proved, with no failures and no Coq expert intervention. The agent writes the proofs under feedback and hard constraints from the harness that keep each one sound (accepted only when the prover's kernel closes it), complete (no obligation left unproved or silently dropped), and terminating (no divergent tactics). We evaluate this harness plus code agent along three dimensions. (1) Core logic: on Iris, the state-of-the-art separation logic for concurrent and memory-manipulating programs, Aria proves all 4,257 lemmas of the four core modules and the 217 lemmas verifying Rust's standard libraries built on it, fully automatically. (2) Comparison with prior LLM provers: on reglang, where prior provers manage barely one in eight, Aria proves all 318. (3) Generality: on iris-lean, the unfinished Lean 4 port of Iris, it proves 72 not-yet-ported lemmas, showing the approach is not specific to Coq. A state-of-the-art model (Claude Opus 4.7) can write proofs for verified software development fully and automatically.

This article presents a solution to intercept an agile drone by a team of agile drone carrying catching nets. We formulate the problem as a competitive Multi-Agent Reinforcement Learning (MARL) task. To address the problem of nonstationarity and catastrophic forgetting of agents overfitting to the current opponent strategy, we train the pursuers and the evader using Multi-Agent Proximal Policy Optimization (MAPPO) with Prioritized Fictitious Self Play (PFSP). We train the agents in a high-fidelity simulator using low-level control commands, collective thrust and body rates (CTBR), to achieve agile flights for both the pursuers and the evader. We compare the performance of the trained policies in terms of catch rate, time to catch and crash rates, against heuristic baselines and show that our solution outperforms them. Ablation studies show that PFSP lead to more robust policies that can adapt to different opponent strategies, and that a low-level control commands are crucial for learning performing strategies in the pursuit-evasion task. Finally, a qualitative analysis of the learned behaviours highlights the emergence of cooperative tactics among the pursuers.

Large Language Model (LLM) agent frameworks such as LangChain, LlamaIndex, and CrewAI have become critical infrastructure powering production AI systems, yet they remain severely under-tested due to fundamental challenges in automated testing. Unlike traditional software, where crashes serve as reliable oracles, defects in these pure Python frameworks manifest as ordinary exceptions or silent semantic failures, creating profound oracle ambiguity. This problem is exacerbated by strict type governance through Pydantic schemas and complex protocol requirements that cause existing fuzzers to generate overwhelming invalid inputs, while traditional test generators produce only trivial cases with weak regression assertions. We present LogicHunter, a fuzzing framework that addresses both the generation and oracle challenges through active specification-aware testing. LogicHunter employs specification-driven generation that systematically fuses formal type constraints with authentic usage patterns from real-world repositories, synthesizing inputs that are valid by construction yet semantically extreme, equipped with behavioral probes to expose silent failures. To resolve oracle ambiguity, we introduce the Agentic Oracle, which transcends passive classification by actively retrieving documentation, navigating source code, and inspecting runtime states through a ReAct-based architecture with Dual-Layer State Management and Dual-Stream Memory. Evaluated on three widely deployed frameworks, LogicHunter discovered 40 previously unknown bugs with 30 confirmed and 26 fixed by developers, while state-of-the-art baselines reported no bugs as final findings. The Agentic Oracle achieves 91.17% precision, surpassing the best passive approach at 29.27% by 61 percentage points.

This paper offers a toy framework for considering curiosity as an ecosystem. First, it suggests that a single agent's inquiry policy (how, when, and why an agent asks a question) depends on how the agent values immediate uncertainty reduction, costs, delayed return, and the value of keeping the question open. A key concept in the framework is that the weights on these decision-related terms can change with experience. For example, a period of cheap, quickly answered questions may change the cost of inquiry on a short timescale and change which kinds of questions the agent is drawn to answer over a longer timescale. Second, these ideas are extended to many agents exploring a shared knowledge landscape, and there the framework tracks inquiry volume, topic diversity, frontier-directed inquiry, redundancy, and reusable knowledge. The result is a conceptual toy framework for studying curiosity ecology and for future efforts towards designing multi-agent AI systems for discovery. It serves as a companion piece for a paper currently under review in Trends in Neurosciences.

LLM-powered data agents are playing an increasingly important role in data-driven decision making. However, existing data agents struggle to generalize to unseen data environments and analytical workflows, especially in heterogeneous enterprise settings. This creates a growing need for synthesizing high-quality data agent trajectories that capture complex analytical workflows for given data environments. Such trajectories support two key downstream uses: they can serve as supervised finetuning (SFT) data that adapts data agent models to the target domain, and as in-context learning (ICL) demonstrations to guide general-purpose LLMs in unfamiliar data environments. Thus, we introduce TOFFEE, a system for synthesizing high-quality data agent trajectories from given data environments via Monte Carlo Tree Search (MCTS) with adaptive model selection and cross-task prefix reuse. We show that TOFFEE can effectively generate scalable trajectory data for complex analytical tasks across heterogeneous environments. In this demonstration, we present the system framework of TOFFEE, including its task pool construction, trajectory explorer, and learned cost model. We also introduce the web interface of TOFFEE and its workflow, and demonstrate two end-to-end scenarios: trajectory synthesis for data agent finetuning, and demonstration-augmented data agent reasoning.

Biomedical question answering requires not only accurate extraction of information from scientific literature but also reliable integration of evidence across multiple documents. This study presents a question-type-specific large language model (LLM) framework for BioASQ 14b Task B, designed to improve answer robustness and evidence grounding in biomedical question answering. Rather than applying a single prompting strategy to all questions, the framework selects different inference procedures for yes/no, factoid, and list questions according to their distinct reasoning and evaluation requirements. For yes/no questions, snippet shuffling and self-reflection are used to reduce sensitivity to evidence ordering and improve decision stability. For factoid questions, full-snippet input is combined with chain-of-thought-based in-context learning to support accurate biomedical entity identification. For list questions, a multi-agent architecture is employed, in which evidence extraction, candidate generation, answer verification, and final aggregation are handled collaboratively. Preliminary experiments on BioASQ 13b were used to identify effective inference strategies for each question type, and the resulting framework was subsequently evaluated in the official BioASQ 14b Task B challenge. In the official evaluation, our framework showed competitive performance across multiple batches and achieved first place in the factoid subtask of Batch 4. These results demonstrate the effectiveness of combining question-type-specific inference, ensemble prediction, and agent-based verification for reliable biomedical question answering.

As web agents increasingly demonstrate capabilities in automated task execution, the development of robust evaluation frameworks for assessing their navigation and task completion performance has emerged as a critical research priority. However, existing benchmarks exhibit fundamental limitations. First, they suffer from insufficient scale and limited domain diversity, constraining comprehensive evaluation of cross-domain generalization. Second, prevailing LLM-as-Judge evaluation methodologies inadequately capture fine-grained interaction semantics, particularly regarding precise query formulation and filtering operations. Third, current benchmarks predominantly emphasize navigation success metrics while neglecting critical requirements for real-world deployment scenarios. To address these limitations, we introduce WebRetriever, a large-scale benchmark encompassing 800 websites and 1,550 tasks across diverse domains, including consumer, professional, and enterprise sectors, with comprehensive coverage of user intent patterns. We propose NavEval (Navigation Evaluation), a novel LLM-as-Judge framework that leverages rich interaction context beyond visual screenshots, achieving state-of-the-art alignment with human judgment across multiple evaluation datasets. Furthermore, we establish three complementary evaluation protocols that collectively provide holistic assessment of web agent capabilities: navigation proficiency, knowledge-assisted interaction, and end-to-end task completion with information extraction. Extensive experimental analysis reveals substantial performance disparities across evaluation protocols, demonstrating that navigation success alone is an insufficient predictor of real-world application effectiveness. WebRetriever delivers fine-grained diagnostic insights into agent capabilities and establishes a rigorous foundation for advancing web agent research and development.

Recent LLM-based mathematical reasoning agents have begun to tackle research-level problems and, in several cases, have contributed to the resolution of open problems. However, scaling and orchestrating such agents effectively remains challenging, due to the difficulty of coordinating parallel proof search while keeping intermediate claims organized and reliable. In this paper, we propose Danus, an orchestration system for research-level mathematical reasoning centered on a shared fact graph as a global memory-management mechanism. Danus consists of a main agent that performs planning and coordination, multiple worker agents that carry out proof search in parallel, and a stateless verifier that checks proposed mathematical claims before they are admitted into the fact graph. Each verified fact is stored together with its proof and logical dependencies, allowing the system to build long arguments incrementally while keeping the shared proof state organized. The main agent periodically summarizes the evolving proof state, redirects workers across promising directions, and supports interaction with human mathematicians through progress reports. We evaluate Danus through six research-level case studies in algebraic geometry, singularity theory, and combinatorics, illustrating how the fact-graph memory mechanism enables Danus to construct long, detailed mathematical proofs. Our results suggest that fact-graph-based orchestration provides an effective route toward scaling mathematical reasoning agents for long-horizon research problems. Danus is open source at https://github.com/frenzymath/Danus.

Multi-agent active visual triangulation enables precise 3D localization of aerial targets by coordinating mobile observers with controllable cameras. However, existing methods assume instantaneous state feedback, ignoring cumulative latency from detection, communication, and decision propagation. We present a delay-aware, uncertainty-driven multi-agent reinforcement learning framework for target localization in Counter-UAS applications. Our contributions are: (1) a Dec-POMDP formulation with Age-of-Information (AoI) augmented observations enabling delay-aware coordination — AoI improves triangulation validity by 10.6 percentage points; (2) a controlled comparison showing that perception-consistent rewards outperform privileged clean-state rewards (0.547 m vs.0.633 m RMSE, 27% fewer track losses) — both policies are trained through identical observation noise but differ in what they are optimized for, producing a stability-robustness tradeoff; and (3) multi-source analytical covariance propagation incorporating pixel, pose, gimbal, and intrinsics uncertainties — restricting to angular noise alone causes 2.8-fold RMSE degradation. Experiments with MAPPO in 4096 parallel environments achieve 0.547 +- 0.217 m RMSE with 78.1% triangulation validity, while MLP policies achieve near-zero validity (0.7%), confirming recurrent memory as essential for delay compensation.

Dilution refrigerators are the enabling infrastructure of superconducting quantum computers, yet their fault diagnosis is still dominated by threshold alarms that report that something is wrong, not what. We present Onnes, a physics-grounded digital-twin simulator of a dilution refrigerator (a forward physics model with a learned real-fridge noise fingerprint) that drives a live multi-agent LLM operations layer, and use it for a controlled head-to-head between a zero-shot LLM agent panel and a supervised ML classifier on cryogenic fault diagnosis. The twin couples a real dilution-cooling floor, a noise-and-correlation fingerprint learned from real BlueFors logs, and six physics-grounded fault classes, three engineered to overlap on temperature but separate on flow and pressure. Across a 1000-turn evaluation the zero-shot panel shows no significant difference from the classifier on detection but trails on classification, its errors concentrating on the confusable faults. Curated contrastive few-shot demonstrations and self-consistency voting then raise classification accuracy from 0.685 to 0.990, matching the supervised classifier (0.985) with no parameter updates and six labeled demonstrations; an ablation attributes the gain almost entirely to the demonstrations. Run as a continuous monitor across a nine-run fault-by-seed sweep, the agent catches every developing fault within one poll interval, and a confidence gate suppresses pre-onset false alarms whose rate is backend-dependent. As a first sim-to-real check, a detector trained purely on real BlueFors telemetry posts a real-hardware false-alarm rate of 6.4% and 100% recall on physics faults injected onto real held-out windows. All numbers are drawn verbatim from released run logs.

Large language model (LLM) agents are increasingly used for multi-step, stateful tool-use tasks, yet production reliability remains limited. Unlike static software repair, agent repair must recover dynamic trajectories whose early decisions can propagate into later errors and external state changes. Existing automatic remedies address only part of this problem: blind retry adds no diagnosis, outcome feedback says whether a run failed but not where or why, and self-reflection often lacks grounded evidence to prevent the same failure from recurring. We present AgentTether, a run-time repair framework that automates post-run diagnosis and guided recovery without modifying the underlying agent or environment. AgentTether abstracts each run into Transition Units, links them through a dependency-aware Critical Transition Graph, and localizes failure-critical subtrajectories by combining an offline normal-behavior model with a run-local graph detector. It then converts the localized cause into behavior-scoped guidance backed by cross-iteration Repair Memory, and can optionally apply guarded run-time intervention to keep the correction active during re-execution. The same design can be deployed as an offline diagnostic-and-guidance tool or as an online repair layer. We evaluate AgentTether on 261 tau-bench tasks across three domains with Qwen3.7-max, and test cross-model transfer on Banking with GPT-5.4. On the hardest Banking domain, AgentTether repairs 59.04% (49/83) of initially failed Qwen3.7-max tasks and 65.12% (56/86) of initially failed GPT-5.4 tasks. Overall, AgentTether improves repair effectiveness while reducing agent turns and end-to-end approach tokens, suggesting a practical reliability layer that can wrap existing agent deployments, reduce wasted re-execution, and improve recovery without retraining the agent.

Autonomous Driving Systems (ADS) can fail because of faults within individual modules as well as from interactions across perception, planning, and control. Yet existing ADS testing research often treats key testing functions, such as perturbation generation, behavioural assessment, and test case selection and exploration, as loosely coupled steps rather than coordinated roles for discovering such failures. We present CREAD, a collaborative multi-agent testing framework for testing ADS that organises perturbation generation, behavioural validation, and search coordination through a shared blackboard and an orchestrator. In the current work-in-progress instantiation, the framework focuses on perception-oriented perturbation generation, while remaining extensible to other ADS modules, including planning and control. It currently comprises a Perception Fuzzer Agent, a Metamorphic Validator Agent, and an Orchestrator Agent. Respectively, they generate perturbations, assess behavioural consistency across related scenario pairs, and coordinate further exploration. Experiments in HighwayEnv simulator show that the collaborative configuration improves failure discovery in the highway environment and remains competitive in the roundabout setting. Across the two environments, it yields about 2.1x as many failures per 100 scenarios as the single-agent baseline on average, while gains over a non-collaborative two-agent baseline vary across environments. These results suggest that collaborative multi-agent testing is a promising research direction for emergent ADS behaviour discovery.

Large language model (LLM) agents have shown strong performance in long-horizon tasks that require planning, tool use, and interaction with external environments. However, most existing benchmarks implicitly assume a monolingual setting, where the entire execution process, including reasoning, tool invocation, and output generation, is conducted within a single language. In contrast, real-world applications often involve multilingual inputs and outputs within a unified workflow, yet the interaction between multilinguality and agentic execution remains underexplored. In this work, we introduce PolyWorkBench, a benchmark for evaluating LLM agents on multilingual long-horizon workplace workflows. PolyWorkBench consists of 67 tasks across five domains, including commerce, knowledge work, legal analysis, localization, and manufacturing, where agents must process heterogeneous multilingual inputs, perform iterative reasoning, invoke external tools, and produce structured outputs. To enable comprehensive evaluation, we propose a hybrid framework that combines structural grading, executable verification, and LLM-based semantic assessment. This design allows us to capture both functional correctness and linguistic consistency across complex workflows. Empirical results show that state-of-the-art LLM agents suffer significant performance degradation in multilingual workflow settings compared to monolingual counterparts. Our analysis suggests that multilinguality introduces compounding effects across reasoning and execution steps, highlighting the importance of jointly modeling language variation and procedural decision-making in agent evaluation.

Large language model (LLM) agents are increasingly evaluated on their ability to use tools, plan multi-step tasks, coordinate with other agents, and operate over extended horizons. Reported benchmark gains often obscure recurring failure modes documented across otherwise unrelated evaluation efforts. This paper synthesizes 27 benchmark, taxonomy, and audit papers (2023-2026), spanning 19 distinct benchmarks, into a cross-cutting taxonomy of agent limitations. To our knowledge, this is the first synthesis that integrates evidence across tool use, planning, long-horizon reasoning, multi-agent coordination, safety, and measurement validity into a single, unified taxonomy of LLM agent limitations. We identify six failure clusters: (1) tool invocation and parameter-level errors, (2) planning and constraint-satisfaction failures, (3) long-horizon degradation from context accumulation, (4) multi-agent coordination failures, (5) safety and security failures under adversarial or underspecified conditions, and (6) measurement validity problems. The taxonomy was derived iteratively by grouping independently reported error categories into themes corresponding to distinct stages of the agent reasoning-to-action pipeline. Across the literature, we find that failures compound nonlinearly with task length, that strong performance on individual sub-tasks does not reliably translate into end-to-end success, and that additional scaffolding does not consistently improve reliability. At the same time, substantial progress has been demonstrated in single-turn tool use, short-horizon web navigation, and narrowly scoped coding tasks.

Recent LLM-based agent systems continuously accumulate context across multi-turn interactions, tool invocations, and cross-session workflows. Replaying the full history for every request quickly becomes impractical: long contexts increase prefill cost, may exceed context limits, and often bury task-relevant evidence in irrelevant content, degrading both serving efficiency and output quality. We propose Akashic, a low-overhead memory system built around MemAttention, which organizes context into bounded chunks and models semantic relationships across chunks, preserving cross-chunk evidence without repeatedly rewriting the full history. Akashic further applies hardware-software co-designed memory placement to co-locate likely co-retrieved chunks, reducing retrieval fragmentation and I/O overhead. Across four representative workloads and three model sizes, Akashic improves task accuracy by up to 10.2 points, throughput by up to 1.21x, and sustainable request rate by up to 1.88x over strong prior memory baselines.

Personal AI agents powered by large language models can reason and act using available tools to access emails, manage calendars, and push code to remote repositories, all with minimal oversight. When augmented with long-term memory, an agent can recall specific details relevant to the current task, reducing the need for large context windows. Currently, long-term memory agents tend to fall into two distinct domains: conversational and action-planning agents. Personal assistant agents sit at the convergence of these two domains and handle sensitive information while interacting with untrusted information sources, creating previously unaccounted security vulnerabilities. In this work, we introduce the novel attack vector, GhostWriter, which exploits current memory subsystems in tool-using personal agents to poison their memory store. GhostWriter operates in two phases: injection, where an adversary sends a hidden attack payload to the target agent; and activation, in which the poisoned memory is retrieved. We show that GhostWriter achieves near-universal injection rates of approximately 98% and a high average activation rate of approximately 60% against state-of-the-art agents. This attack is possible due to the lack of security-focused memory governance. In response, we propose Agentic Memory Sentry (AM-Sentry), which leverages two mitigation techniques: a memory-saving policy and a memory-retrieval screen. Our experiments show that AM-Sentry dramatically reduces GhostWriter's success rate while preserving agent utility.