Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
arXiv:2611.05827v1 Announce Type: new Abstract: Decision models choose among predefined options in tasks such as routing and action selection. In multi-agent collaboration, a decision determines how...
arXiv:2611.06007v1 Announce Type: new Abstract: Motorized Two-Wheelers (MTWs) dominate personal transport across much of the Global South, yet remain largely overlooked in predictive perception resea...
arXiv:2611.05398v1 Announce Type: cross Abstract: Colony counting is commonly used to evaluate microbial populations in food safety testing and microbiological studies. Despite its wide application,...
arXiv:2602.11499v3 Announce Type: replace Abstract: Multimodal Large Language Models have shown promising capabilities in bridging visual and textual reasoning, yet their reasoning capabilities in Op...
arXiv:2605.20342v3 Announce Type: replace Abstract: Training large multimodal models (LMMs) via reinforcement learning (RL) to natively invoke video-processing tools (e.g., cropping) has become a pro...
arXiv:2611.00180v2 Announce Type: replace-cross Abstract: Embodied agents are usually organized around individual tasks, yet the physical world keeps changing between tasks and during execution. We i...
The study evaluates how agentic AI models can autonomously complete end‑to‑end data science projects and improve through iterative feedback. A writer model generates a full project—from data acquisition to analysis and paper writing—while a grader model evaluates it; this cycle repeats for up to three rounds. Across 49 writer‑grader pairs from five developers, initial submissions improved from an average of 59 % to 76 % after feedback, with variations depending on the specific model roles.
PAYWORTHY is a new benchmark that tests whether large language models can predict the success of an AI agent on freelance jobs before the agent attempts them. The dataset contains 1,494 real job postings sampled from 18.8 million postings on a freelancing platform, and 18 models are asked to forecast the likelihood of success. When the agent’s limitations are explicitly described, models achieve AUC scores between 0.66 and 0.86; when the agent is described generically, the models perform near random (AUC 0.45–0.56).
The paper introduces an autonomous planning framework for electricity systems that iteratively proposes, evaluates, and refines investment and operational decisions across regional scenarios. Agents generate investment bounds, storage conditions, and feeder interventions, interpret numerical feedback, and decide when to stop, automatically running annual investment-and-dispatch optimization and feeder verification. The framework was tested on forty carbon‑constrained trajectories across five systems and four countries, successfully retaining low‑cost candidates that meet physical and carbon‑budget constraints, with 28 of 40 matching or exceeding reference designs.
ModalSearcher introduces an autonomous cross‑modal search framework that lets an agent dynamically choose among direct reasoning, text‑to‑text search, text‑to‑image search, and reverse‑image search, actively incorporating visual evidence as needed. The system builds an interleaved multimodal knowledge graph via a scalable pipeline that samples diverse paths and uses backward question construction to prevent shortcut reasoning. Trained with supervised fine‑tuning and reinforcement learning, ModalSearcher‑8B outperforms Qwen3‑VL‑8B by 16.2 accuracy points across five multimodal knowledge‑intensive QA benchmarks and shows consistent gains on active visual‑search tasks.
The paper proposes a temporal organization layer for continual agents, introducing phase‑tagged micro‑epochs, a duty cycle, and nested clocks to separate exploration from commitment. It presents deterministic and stochastic guarantees for duty‑cycle updates, defines busy/idle execution contracts, and evaluates the design through synthetic experiments and a Qwen3‑8B prototype. The results demonstrate a testable scheduling design and clarify its conditions of validity.
The paper argues that large language model agents, which persist across sessions and accumulate memory, suffer from unchecked plasticity leading to semantic drift, retrieval errors, and memory saturation. It draws on a selectionist theory of sleep—comprising ringing, resonance, and renormalization—to propose that each phase can be implemented in digital memory systems. The authors predict that selectionist consolidation will outperform instructionist methods as memory stores grow, suggesting that sleep is a necessary cost of continuous learning under finite capacity.
BobFlow proposes using dataflows as a shared computational abstraction to facilitate human-agent collaboration in data science, combining this with the ReAct paradigm to allow agents to iteratively build and run dataflows while analysts can inspect and provide feedback. The system demonstrates that analysts can better understand, inspect, and verify agent-generated analyses, and experimental results show accuracy gains of up to 18.6 points at roughly one‑third the inference cost of a script‑based ReAct agent.
The paper introduces GREP, a framework for evaluating AI-generated furnished rooms that moves beyond specific questions to a three-tier philosophy: basic physical validity, visual coherence, and human-use logic. An autonomous evaluation agent applies these principles to identify scene issues, produce scores, and generate reports, enabling reliable, scene-specific judgments. Experiments demonstrate GREP’s high agreement with human preferences, its ability to improve scene quality through feedback, and insights into how future models might better balance planning and constraint adaptation.
The article presents a reference framework for autonomous extract‑transform‑load (ETL) pipelines to consolidate Salesforce data during mergers and acquisitions. It details a partitioned bulk extraction, declarative mapping, rule‑based and statistical quality gates, master data management with deterministic and blocked fuzzy matching, constrained loading with error capture, and change data capture protected by a data‑health circuit breaker. The prototype was evaluated on synthetic scenarios, achieving high F1 scores for account and contact matching and preventing load failures and data loss.
The paper investigates zero‑shot coordination (ZSC) among large language model (LLM) agents, a scenario where independently developed agents must coordinate without prior shared scaffolding. Experiments in minimal, controlled environments reveal that current frontier LLMs struggle to coordinate, mainly due to limited understanding of the coordination problem and weak reasoning about partners’ beliefs. The authors introduce Coordination‑Friendly Definitions (CFDs) as a principled, automatically discoverable approach to improve robust coordination among LLM agents.
EvoWorld introduces a framework that couples a vision‑language model with a segment‑autoregressive video renderer to enable generative video models to reason about world evolution beyond visual fidelity. The system maintains an evolving account of the world, infers event consequences, and renders causally plausible outcomes even when events occur out of sight. The accompanying EvoWorld‑Data and EvoWorld‑Bench datasets provide training and evaluation for event realization, causal reactions, and revisit consistency.
The paper studies a repeated principal–agent game where the principal cannot observe the agent’s rewards and both parties learn over time. It introduces a non‑parametric estimator that uses only the history of incentives and choices, and combines it with a data‑driven incentive policy in a multi‑armed bandit setting. The authors prove finite‑sample consistency of the estimator and provide a regret bound for the principal, supported by simulations on green‑energy aggregator contracts.
The paper introduces a visual-native agent harness that registers every tool‑returned image as an addressable reference, enabling reuse of intermediate visual evidence in multimodal deep search tasks. It also proposes On‑Policy Data Evolution (ODE), a closed‑loop data generator that refines training data across rounds based on policy rollouts, targeting the agent’s evolving needs. Experiments on eight benchmarks show ODE boosts the Qwen3‑VL‑8B agent’s performance from 24.9% to 39.0% on average, and from 30.6% to 41.5% at 30B, while validating the benefits of image bank reuse and rollout‑feedback evolution.
SlackBench is a new benchmark that evaluates AI agents on realistic collaborative research and engineering workflows, requiring joint reasoning over workplace communication and evolving repository state. The benchmark is built from privacy‑preserving reenactments of real academic projects, providing reproducible environments with multi‑party dialogue, code changes from GitHub, and Google Docs. It contains 140 queries covering heterogeneous‑source search, abstention, summarization, and inference over long conversations, and shows that current models perform well below the ceiling, indicating significant room for improvement.