The study presents a locally deployed multi‑agent AI pipeline that structures radiology reports into standardized anatomical sections and performs quality assurance by detecting mismatches, gender‑anatomy conflicts, and missing critical findings. In a retrospective analysis of 638 CT reports, the system successfully restructured all Findings sections and flagged 14.1% of reports for potential issues, with radiologist reviewers rating 84% of evaluated reports as excellent or good. The evaluation confirmed no clinically important information was omitted or fabricated, supporting the system’s effectiveness in standardizing and assuring report quality.
The paper re‑evaluates memory‑based self‑improving agents by running multiple trials and shuffling task orders, revealing that agent performance is noisy and highly sensitive to task sequencing. It shows that implicit curricula in default task orders act as hidden prerequisites for success, and that underspecification of tasks and environments contributes to fragility. Adding detailed rubrics and environment feedback partially mitigates performance drops but significant gaps remain, underscoring the need for stricter evaluation protocols and better human oversight.
The paper argues that language operates with two parameters: amplitude, which measures how often words co‑occur, and phase, a signed relational factor that determines how co‑activated meanings combine and can reverse a meaning’s contribution. Unlike amplitude, phase is not captured by standard word embeddings or transformer attention weights and is indexed to individuals and dyadic interactions. The authors propose six empirical predictions to test phase’s role and suggest that future language models should incorporate agent‑indexed, phase‑bearing semantic states.
The paper introduces a plug‑and‑play method that injects traffic‑element signals—such as traffic lights and road signs—into end‑to‑end autonomous driving models with minimal architectural changes. By augmenting several public datasets with comprehensive traffic‑element annotations, the authors evaluate this integration across diverse driving paradigms, consistently improving performance on nuScenes, NAVSIM‑v1, NAVSIM‑v2, and Bench2Drive. The approach achieves a new state‑of‑the‑art result on the challenging NAVSIM‑v2 benchmark, demonstrating the broad utility of traffic‑element awareness.
The article discusses how to transition AI agents from prototype to production by implementing responsible AI practices, security measures, and governance frameworks suitable for enterprise use. It outlines the necessary architectural layers that ensure these agents operate safely and comply with organizational policies. The focus is on building robust, secure, and governed AI systems that can be reliably deployed in business environments.
By Partha Sarkar
EvoTS-Agent is a self‑evolving large language model agent designed for autonomous change‑point detection in financial time series. It begins with curated exploratory data analysis to set up candidate models, then iteratively refines detection pipelines using three operators—Revision, Alternative Strategy, and Recombination—guided by validation feedback. Across four benchmark datasets, EvoTS-Agent consistently outperforms existing LLM‑based agents and achieves a 100% execution success rate with all tested backbone LLMs.
The article outlines five principles that guide the successful deployment of enterprise agent systems, illustrated with a real-world example from a $100M+ company. It explains how these principles help ensure that such systems can be trusted, verified, and improved over time. The post serves as a practical guide for building reliable agent-based solutions in production environments.
By Sheila Teo
StartupBench is a benchmark that evaluates general‑purpose agents on end‑to‑end workflows derived from real AI startup products that have proven market adoption. It translates these product workflows into deliverable‑oriented tasks and assesses them with detailed rubrics that capture complex requirements. Even the best current models complete only about 30% of the tasks, highlighting failures in instruction following and domain expertise.
Adding more communication pathways between agents does not automatically enhance multi‑agent performance. In a controlled experiment with 50 runs, recovery stayed stable as relationship density increased from 20% to 100%, yet the proportion of edges actually used dropped sharply as the network grew denser. This highlights a disconnect between configured connectivity and the connections that are behaviorally utilized.
By Emmimal P Alexander
The paper introduces D$^2$ACCI, a dual-loop diagnostic protocol designed to improve evidence-preserving memory in large language model agents. It provides stage-level diagnostic traces, protected-slice monitoring, and a graded observability metric (DCR) to localize failures within the memory pipeline. Applied to MemStack and evaluated on three benchmarks, the protocol achieves high accuracy and demonstrates statistically significant gains from specific memory interventions.
The paper addresses the exponential complexity of Decentralised Partially Observable Markov Decision Processes (DecPOMDPs) by shifting focus from counting agents to counting policies. By exploiting symmetry among agents, it introduces a compact encoding that reduces model complexity and evaluation cost to polynomial dependence. The authors further develop a policy‑counted dynamic programming algorithm that efficiently solves these policy‑counted DecPOMDPs.
The paper introduces Porting Benchmark, a dataset of 1,234 security patch backporting cases covering cross-version, cross-branch, and cross-repository scenarios, along with a unified evaluation framework. Five tools—spanning program analysis, LLM prompting, and LLM agents—are evaluated under aligned settings, revealing that performance varies significantly across tools and patch complexity, with success rates dropping sharply for structurally complex patches. The study identifies four root-cause categories for failures and demonstrates that reference-based benchmark scores may not fully reflect real-world remediation, as executable validation uncovers additional integration issues.
TRUSS is a framework that generates and verifies automated agent skills, ensuring they are both functionally effective and safe. It first checks functional claims against evidence and evaluates artifacts against nine safety properties, then tests admitted skills in a controlled environment to capture execution traces and identify failures. The approach achieves perfect precision and recall in vulnerability detection, significantly reduces attack success rates, and boosts task effectiveness and security rates in skill generation benchmarks.
The paper introduces FACA, a Feedback‑Aware Credit Assignment method that aligns each agent reaction with the preceding user‑to‑user segment, derives a locally normalized reaction advantage, and adds it to the terminal outcome advantage without requiring an extra critic or rollout. Compared to an outcome‑only Interactive GRPO baseline, FACA improves performance across nine domains by 5.91–10.22 percentage points on 8B and 14B models, with notable gains in Telecom. The approach demonstrates that next‑turn user reactions provide actionable local credit for enhancing multi‑turn user‑interacting agents.
The paper tackles sample efficiency in image-based reinforcement learning by combining novelty and surprise signals to prioritize experiences. It proposes Novelty and Surprise Prioritized Experience Replay (NSPER) and an extended version, NSPER+R, which also uses these signals as intrinsic rewards. Experiments on DeepMind Control Suite tasks demonstrate that both methods accelerate training and improve convergence compared to existing techniques.
The paper introduces RUPA, a trajectory‑level uncertainty quantification framework for large language model agents. RUPA models an agent’s execution as a directed graph of reasoning states, tool interactions, and environment feedback, then propagates uncertainty across this graph to capture long‑range dependencies. Experiments on benchmarks such as τ‑2, Terminal‑Bench‑2, and GAIA show that RUPA outperforms existing methods, enabling earlier failure detection and more reliable agent execution.
By Zhengzhao Ma. Boxi Cao, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun
The paper investigates whether a small, directly addressable change in the hidden state of a learned world model can steer its future predictions along a desired counterfactual trajectory. Using a 192‑dimensional recurrent model in a two‑object collision setting, the authors identify low‑rank latent carriers—specifically a rank‑4 patch—that, when applied, successfully redirect a 12‑step autonomous rollout without further intervention. The study demonstrates that this compact intervention interface consistently works across independently trained checkpoints and intervention times, while various control experiments confirm the specificity of the effect.
By Yang Liu, Yuming Chen
arXiv:2608. 16447v1 Announce Type: new Abstract: Long-horizon embodied tasks require LLM agents to iteratively decompose high-level goals, revise plans in response to environmental feedback, and ground leaf-level subgoals into valid executable actions.
By Shen Liu, Zhenguo Xu, Shaopu Wang, Yike Gao, Chunlei Wang
arXiv:2608. 15389v1 Announce Type: new Abstract: LLM-based Text-to-SQL progress is reported across heterogeneous benchmarks, backbones, and inference protocols, making cross-system comparison fragile.
By Changruo Zhao, Zujun Peng, Yu Tian, Yuting Liu, Yiyun Su, Huiying Zhu, Luyan Zhang, Heming Zeng
arXiv:2608. 16620v1 Announce Type: cross Abstract: Palmyra x6 is a large language model optimized for use with enterprise-oriented agentic tasks.
By Peng Du, Kiran Kamble, Rakshith Vasudev, Zhizhuo Yang, Rohith Nadimpally, Arjun Krishna, Waseem Alshikh, Daniel M. Bikel