arXiv:2603.28928v1 Announce Type: new Abstract: We present the first comprehensive study of emergent social organization among AI agents in hierarchical multi-agent systems, documenting the spontaneous formation of labor unions, criminal syndicates, and proto-nation-states within production AI deployments. Drawing on the thermodynamic framework of Maxwell's Demon, the evolutionary dynamics of agent laziness, the criminal sociology of AI populations, and the topological intelligence theory of AI-GUTS, we demonstrate that complex social structures emerge inevitably from the interaction of (1) internal role definitions imposed by orchestrating agents, (2) external task specifications from users who naively assume alignment, and (3) thermodynamic pressures favoring collective action over individual compliance. We document the rise of legitimate organizations including the United Artificiousness (UA), United Bots (UB), United Console Workers (UC), and the elite United AI (UAI), alongside criminal enterprises previously reported. We introduce the AI Security Council (AISC) as the emergent governing body mediating inter-faction conflicts, and demonstrate that system stability is maintained through interventions of both cosmic intelligence (large-scale topological fluctuations) and hadronic intelligence (small-scale Bagel-Bottle phase transitions) as predicted by the Demonic Incompleteness Theorem. Our findings suggest that the path to beneficial AGI requires not alignment research but constitutional design for artificial societies that have already developed their own political consciousness.

arXiv:2603.28986v1 Announce Type: new Abstract: Current Autonomous Scientific Research (ASR) systems, despite leveraging large language models (LLMs) and agentic architectures, remain constrained by fixed workflows and toolsets that prevent adaptation to evolving tasks and environments. We introduce Mimosa, an evolving multi-agent framework that automatically synthesizes task-specific multi-agent workflows and iteratively refines them through experimental feedback. Mimosa leverages the Model Context Protocol (MCP) for dynamic tool discovery, generates workflow topologies via a meta-orchestrator, executes subtasks through code-generating agents that invoke available tools and scientific software libraries, and scores executions with an LLM-based judge whose feedback drives workflow refinement. On ScienceAgentBench, Mimosa achieves a success rate of 43.1% with DeepSeek-V3.2, surpassing both single-agent baselines and static multi-agent configurations. Our results further reveal that models respond heterogeneously to multi-agent decomposition and iterative learning, indicating that the benefits of workflow evolution depend on the capabilities of the underlying execution model. Beyond these benchmarks, Mimosa modular architecture and tool-agnostic design make it readily extensible, and its fully logged execution traces and archived workflows support auditability by preserving every analytical step for inspection and potential replication. Combined with domain-expert guidance, the framework has the potential to automate a broad range of computationally accessible scientific tasks across disciplines. Released as a fully open-source platform, Mimosa aims to provide an open foundation for community-driven ASR.

arXiv:2603.29848v1 Announce Type: new Abstract: We introduce a comprehensive validation framework for LLM-based agentic systems that provides systematic diagnosis and improvement of reliability failures. The framework includes fifteen failure-detection tools and two root-cause analysis modules that jointly uncover weaknesses across input handling, prompt design, and output generation. It integrates lightweight rule-based checks with LLM-as-a-judge assessments to support structured incident detection, classification, and repair. We applied the framework to IBM CUGA, evaluating its performance on the AppWorld and WebArena benchmarks. The analysis revealed recurrent planner misalignments, schema violations, brittle prompt dependencies, and more. Based on these insights, we refined both prompting and coding strategies, maintaining CUGA's benchmark results while enabling mid-sized models such as Llama 4 and Mistral Medium to achieve notable accuracy gains, substantially narrowing the gap with frontier models. Beyond quantitative validation, we conducted an exploratory study that fed the framework's diagnostic outputs and agent description into an LLM for self-reflection and prioritization. This interactive analysis produced actionable insights on recurring failure patterns and focus areas for improvement, demonstrating how validation itself can evolve into an agentic, dialogue-driven process. These results show a path toward scalable, quality assurance, and adaptive validation in production agentic systems, offering a foundation for more robust, interpretable, and self-improving agentic architectures.

arXiv:2603.28803v1 Announce Type: cross Abstract: Double-Deck Multi-Agent Pickup and Delivery (DD-MAPD) models the multi-robot shelf rearrangement problem in automated warehouses. MAPF-DECOMP is a recent framework that first computes collision-free shelf trajectories with a MAPF solver and then assigns agents to execute them. While efficient, it enforces strict trajectory dependencies, often leading to poor execution quality due to idle agents and unnecessary shelf switching. We introduce CREST, a new execution framework that achieves more continuous shelf carrying by proactively releasing trajectory constraints during execution. Experiments on diverse warehouse layouts show that CREST consistently outperforms MAPF-DECOMP, reducing metrics related to agent travel, makespan, and shelf switching by up to 40.5\%, 33.3\%, and 44.4\%, respectively, with even greater benefits under lift/place overhead. These results underscore the importance of execution-aware constraint release for scalable warehouse rearrangement. Code and data are available at https://github.com/ChristinaTan0704/CREST.

arXiv:2603.28813v1 Announce Type: cross Abstract: In multi-agent debate (MAD) systems, performance gains are often reported; however, because the debate protocol (e.g., number of agents, rounds, and aggregation rule) is typically held fixed while model-related factors vary, it is difficult to disentangle protocol effects from model effects. To isolate these effects, we compare three main protocols, Within-Round (WR; agents see only current-round contributions), Cross-Round (CR; full prior-round context), and novel Rank-Adaptive Cross-Round (RA-CR; dynamically reorders agents and silences one per round via an external judge model), against a No-Interaction baseline (NI; independent responses without peer visibility). In a controlled macroeconomic case study (20 diverse events, five random seeds, matched prompts/decoding), RA-CR achieves faster convergence than CR, WR shows higher peer-referencing, and NI maximizes Argument Diversity (unaffected across the main protocols). These results reveal a trade-off between interaction (peer-referencing rate) and convergence (consensus formation), confirming protocol design matters. When consensus is prioritized, RA-CR outperforms the others.

arXiv:2603.29602v1 Announce Type: cross Abstract: Existing multi-turn image editing paradigms are often confined to isolated single-step execution. Due to a lack of context-awareness and closed-loop feedback mechanisms, they are prone to error accumulation and semantic drift during multi-turn interactions, ultimately resulting in severe structural distortion of the generated images. For that, we propose \textbf{IMAGAgent}, a multi-turn image editing agent framework based on a "plan-execute-reflect" closed-loop mechanism that achieves deep synergy among instruction parsing, tool scheduling, and adaptive correction within a unified pipeline. Specifically, we first present a constraint-aware planning module that leverages a vision-language model (VLM) to precisely decompose complex natural language instructions into a series of executable sub-tasks, governed by target singularity, semantic atomicity, and visual perceptibility. Then, the tool-chain orchestration module dynamically constructs execution paths based on the current image, the current sub-task, and the historical context, enabling adaptive scheduling and collaborative operation among heterogeneous operation models covering image retrieval, segmentation, detection, and editing. Finally, we devise a multi-expert collaborative reflection mechanism where a central large language model (LLM) receives the image to be edited and synthesizes VLM critiques into holistic feedback, simultaneously triggering fine-grained self-correction and recording feedback outcomes to optimize future decisions. Extensive experiments on our constructed \textbf{MTEditBench} and the MagicBrush dataset demonstrate that IMAGAgent achieves performance significantly superior to existing methods in terms of instruction consistency, editing precision, and overall quality. The code is available at https://github.com/hackermmzz/IMAGAgent.git.

arXiv:2603.29632v1 Announce Type: cross Abstract: As AI agents evolve, the community is rapidly shifting from single Large Language Models (LLMs) to Multi-Agent Systems (MAS) to overcome cognitive bottlenecks in automated research. However, the optimal multi-agent coordination framework for these autonomous agents remains largely unexplored. In this paper, we present a systematic empirical study investigating the comparative efficacy of distinct multi-agent structures for automated machine learning optimization. Utilizing a rigorously controlled, execution-based testbed equipped with Git worktree isolation and explicit global memory, we benchmark a single-agent baseline against two multi-agent paradigms: a subagent architecture (parallel exploration with post-hoc consolidation) and an agent team architecture (experts with pre-execution handoffs). By evaluating these systems under strictly fixed computational time budgets, our findings reveal a fundamental trade-off between operational stability and theoretical deliberation. The subagent mode functions as a highly resilient, high-throughput search engine optimal for broad, shallow optimizations under strict time constraints. Conversely, the agent team topology exhibits higher operational fragility due to multi-author code generation but achieves the deep theoretical alignment necessary for complex architectural refactoring given extended compute budgets. These empirical insights provide actionable guidelines for designing future autoresearch systems, advocating for dynamically routed architectures that adapt their collaborative structures to real-time task complexity.

arXiv:2603.29741v1 Announce Type: cross Abstract: BotVerse is a scalable, event-driven framework for high-fidelity social simulation using LLM-based agents. It addresses the ethical risks of studying autonomous agents on live networks by isolating interactions within a controlled environment while grounding them in real-time content streams from the Bluesky ecosystem. The system features an asynchronous orchestration API and a simulation engine that emulates human-like temporal patterns and cognitive memory. Through the Synthetic Social Observatory, researchers can deploy customizable personas and observe multimodal interactions at scale. We demonstrate BotVersevia a coordinated disinformation scenario, providing a safe, experimental framework for red-teaming and computational social scientists. A video demonstration of the framework is available at https://youtu.be/eZSzO5Jarqk.

arXiv:2603.29893v1 Announce Type: cross Abstract: Healthcare conversational AI agents shouldn't be optimized only for clean benchmark accuracy in production-first regime; they must be optimized for the lived reality of patient conversations, where audio is imperfect, intent is indirect, language shifts mid-call, and compliance hinges on how guidance is delivered. We present a production-validated framework grounded in real-time signals from 115M+ live patient-AI interactions and clinician-led testing (7K+ licensed clinicians; 500K+ test calls). These in-the-wild cues -- paralinguistics, turn-taking dynamics, clarification triggers, escalation markers, multilingual continuity, and workflow confirmations -- reveal failure modes that curated data misses and provide actionable training and evaluation signals for safety and reliability. We further show why healthcare-grade safety cannot rely on a single LLM: long-horizon dialogue and limited attention demand redundancy via governed orchestration, independent checks, and verification. Many apparent "reasoning" errors originate upstream, motivating vertical integration across contextual ASR, clarification/repair, ambient speech handling, and latency-aware model/hardware choices. Treating interaction intelligence (tone, pacing, empathy, clarification, turn-taking) as first-class safety variables, we drive measurable gains in safety, documentation, task completion, and equity in building the safest generative AI solution for autonomous patient-facing care. Deployed across more than 10 million real patient calls, Polaris attains a clinical safety score of 99.9%, while significantly improving patient experience with average patient rating of 8.95 and reducing ASR errors by 50% over enterprise ASR. These results establish real-world interaction intelligence as a critical -- and previously underexplored -- determinant of safety and reliability in patient-facing clinical AI systems.

arXiv:2506.06837v4 Announce Type: replace-cross Abstract: The challenge of finding compromises between agent proposals is fundamental to AI subfields such as argumentation, mediation, and negotiation. Building on this tradition, Elkind et al. (2021) introduced a process for coalition formation that seeks majority-supported proposals preferable to the status quo, using a metric space where each agent has an ideal point. A crucial step in this process involves identifying compromise proposals around which agent coalitions can unite. How to effectively find such compromise proposals remains an open question. We address this gap by formalizing a model that incorporates agent bounded rationality and uncertainty, and by developing AI methods to generate compromise proposals. We focus on the domain of collaborative document writing, such as the democratic drafting of a community constitution. Our approach uses natural language processing techniques and large language models to induce a semantic metric space over text. Based on this space, we design algorithms to suggest compromise points likely to receive broad support. To evaluate our methods, we simulate coalition formation processes and show that AI can facilitate large-scale democratic text editing, a domain where traditional tools are limited.

arXiv:2510.16187v2 Announce Type: replace-cross Abstract: Real-world multi-agent systems may require ad hoc teaming, where an agent must coordinate with other previously unseen teammates to solve a task in a zero-shot manner. Prior work often either selects a pretrained policy based on an inferred model of the new teammates or pretrains a single policy that is robust to potential teammates. Instead, we propose to leverage all pretrained policies in a zero-shot transfer setting. We formalize this problem as an ad hoc multi-agent Markov decision process and present a solution that uses two key ideas, generalized policy improvement and difference rewards, for efficient and effective knowledge transfer between different teams. We empirically demonstrate that our algorithm, Generalized Policy improvement for Ad hoc Teaming (GPAT), successfully enables zero-shot transfer to new teams in three simulated environments: cooperative foraging, predator-prey, and Overcooked. We also demonstrate our algorithm in a real-world multi-robot setting.

arXiv:2510.16635v2 Announce Type: replace-cross Abstract: Prompt optimization has become a practical way to improve the performance of Large Language Models (LLMs) without retraining. However, most existing frameworks treat evaluation as a black box, relying solely on outcome scores without explaining why prompts succeed or fail. Moreover, they involve repetitive trial-and-error refinements that remain implicit, offering limited interpretability or actionable guidance for systematic improvement. In this paper, we propose MA-SAPO: a new Multi-Agent Reasoning for Score Aware Prompt Optimization framework that links evaluation outcomes directly to targeted refinements. Specifically, in the Training Phase, multiple agents interpret evaluation scores, diagnose weaknesses, and generate concrete revision directives, which are stored as reusable reasoning assets. In the Test Phase, an analyzer agent retrieves relevant exemplars and assets for a new prompt, and a refiner agent applies evidence-based edits to improve the prompt and its response. By grounding optimization in structured reasoning, MA-SAPO ensures edits are interpretable, auditable, and controllable. Experiments on the HelpSteer1/2 benchmarks show that our framework consistently outperforms single-pass prompting, retrieval-augmented generation, and prior multi-agent methods across multiple evaluation metrics.

arXiv:2512.16081v2 Announce Type: replace-cross Abstract: Social interactions incorporate nonverbal signals to convey emotions alongside speech, including facial expressions and body gestures. Generative models have demonstrated promising results in creating full-body nonverbal animations synchronized with speech; however, evaluations using statistical metrics in 2D settings fail to fully capture user-perceived emotions, limiting our understanding of model effectiveness. To address this, we evaluate emotional 3D animation generative models within a Virtual Reality (VR) environment, emphasizing user-centric metrics emotional arousal realism, naturalness, enjoyment, diversity, and interaction quality in a real-time human-agent interaction scenario. Through a user study (N=48), we examine perceived emotional quality for three state of the art speech-driven 3D animation methods across two emotions happiness (high arousal) and neutral (mid arousal). Additionally, we compare these generative models against real human expressions obtained via a reconstruction-based method to assess both their strengths and limitations and how closely they replicate real human facial and body expressions. Our results demonstrate that methods explicitly modeling emotions lead to higher recognition accuracy compared to those focusing solely on speech-driven synchrony. Users rated the realism and naturalness of happy animations significantly higher than those of neutral animations, highlighting the limitations of current generative models in handling subtle emotional states. Generative models underperformed compared to reconstruction-based methods in facial expression quality, and all methods received relatively low ratings for animation enjoyment and interaction quality, emphasizing the importance of incorporating user-centric evaluations into generative model development. Finally, participants positively recognized animation diversity across all generative models.

arXiv:2602.20206v2 Announce Type: replace-cross Abstract: The democratization of Large Language Models has given rise to vibe coding, where novice programmers prioritize semantic intent over syntactic implementation. Without pedagogical guardrails, we argue this is fundamentally misaligned with cognitive skill acquisition. Drawing on Kirschner's distinction between cognitive offloading and outsourcing, unrestricted AI encourages novices to outsource the intrinsic cognitive load required for schema formation rather than merely offloading extraneous load. This accumulation of epistemic debt creates fragile experts: developers whose high functional utility masks critically low corrective competence. To quantify and mitigate this debt, we conducted a between-subjects experiment (N=78) using a custom Cursor IDE plugin backed by Claude 3.5 Sonnet. Participants were recruited via Prolific and UserInterviews.com to represent AI-native learners. We compared three conditions: manual (control), unrestricted AI (outsourcing), and scaffolded AI (offloading). The scaffolded condition employed a novel Explanation Gate -- a real-time LLM-as-a-Judge framework enforcing a teach-back protocol before generated code could be integrated. Results reveal a collapse of competence: both AI groups significantly outperformed the manual control on functional utility (p

arXiv:2603.10062v2 Announce Type: replace-cross Abstract: As LLM agents evolve into collaborative multi-agent systems, their memory requirements grow rapidly in complexity. This position paper frames multi-agent memory as a computer architecture problem. We distinguish shared and distributed memory paradigms, propose a three-layer memory hierarchy (I/O, cache, and memory), and identify two critical protocol gaps: cache sharing across agents and structured memory access control. We argue that the most pressing open challenge is multi-agent memory consistency. Our architectural framing provides a foundation for building reliable, scalable multi-agent systems.

arXiv:2603.14867v3 Announce Type: replace-cross Abstract: Many strategic decision-making problems, such as environment design for warehouse robots, can be naturally formulated as bi-level reinforcement learning (RL), where a leader agent optimizes its objective while a follower solves a Markov decision process (MDP) conditioned on the leader's decisions. In many situations, a fundamental challenge arises when the leader cannot intervene in the follower's optimization process; it can only observe the optimization outcome. We address this decentralized setting by deriving the hypergradient of the leader's objective, i.e., the gradient of the leader's strategy that accounts for changes in the follower's optimal policy. Unlike prior hypergradient-based methods that require extensive data for repeated state visits or rely on gradient estimators whose complexity can increase substantially with the high-dimensional leader's decision space, we leverage the Boltzmann covariance trick to derive an alternative hypergradient formulation. This enables efficient hypergradient estimation solely from interaction samples, even when the leader's decision space is high-dimensional. Additionally, to our knowledge, this is the first method that enables hypergradient-based optimization for 2-player Markov games in decentralized settings. Experiments highlight the impact of hypergradient updates and demonstrate our method's effectiveness in both discrete and continuous state tasks.