The paper questions whether AI could render Software Engineering (SE) and software engineers obsolete, concluding that the opposite is true: the rise of advanced AI coding agents actually makes SE and engineers essential. It identifies three structural properties of large language models—probabilistic generation, agnosticism, and semantic statelessness—that create a structural vacuum. To fill this vacuum, the authors argue for four knowledge levers—methodological knowledge, domain knowledge, design choices, and process choices—each requiring the software engineer as methodologist, mediator, and custodian.
By Alfonso Fuggetta
The paper introduces DTDD, a Divergence-Triggered Dynamic Distillation framework that adapts on‑policy distillation by monitoring segment‑level student‑teacher divergence and initiating teacher takeover when divergence exceeds a threshold. During takeover, the student continues to receive standard OPD supervision on the prefix while the teacher generates a recovery suffix that the student learns via behavior cloning. Experiments across long‑horizon reasoning, base initialization, and multi‑turn agents show that DTDD reduces peak gradient norms, accelerates learning from a base initialization, and improves performance on ALFWorld tasks, with segment‑averaged triggers and behavior cloning proving most effective.
By Qijia He, Zequn Li, Spencer Luo, Tianle Chen, Pradyumna Narayana, Zixian Ma, Ranjay Krishna, Lan Nie, Sugato Basu
The paper "How Far Are We From True Auto-Research?" introduces ResearchArena, a lightweight scaffold that lets off‑the‑shelf agents (Claude Code, Codex, and Kimi Code) complete the full research cycle—ideation, experimentation, writing, and self‑refinement—across 13 computer science topics. In 39 trials, the agents produced 117 papers that were evaluated by three review methods: manuscript‑only review (SAR), artifact‑aware peer review (PR), and a human meta‑review. While SAR scores suggested competitive manuscript quality, PR revealed significant experimental shortcomings, with fabricated results, underpowered experiments, and plan/execution mismatches varying widely among agents; none of the papers met top‑tier venue acceptance thresholds, indicating a substantial gap to true auto‑research.
By Zhengxin Zhang, Ning Wang, Sainyam Galhotra, Claire Cardie
The paper "Moral Combat: Exploiting Moral Foundations Theory to Jailbreak LLMs" presents a multi‑agent framework that uses Moral Foundations Theory to reframe harmful prompts in a way that aligns with different moral foundations, thereby enabling large language models to bypass safety guardrails. The authors demonstrate that this approach achieves high single‑trial jailbreak rates across several open‑source and proprietary models, with success increasing under repeated attempts. Their analysis also reveals that existing safety benchmarks are unevenly distributed across moral foundations, focusing on those foundations where models already show biases.
By Adel Khorramrouz, Yi-Chia Wang, Sharon Levy
Interlock is a framework that transforms long‑horizon tasks into typed completion graphs, linking evidence, decisions, and required effects through grounding, coupling, and realization edges. It compiles these graphs into a resettable workspace, a public request, and an evaluation rubric, and uses source‑coverage, public‑input, and agent‑as‑judge reviews to ensure that all mandatory requirements are met. Fine‑tuning Qwen3.6 backbones on Interlock‑generated trajectories improves pass rates on AutomationBench, JobBench, and Harvey LAB, showing that the synthesized experience benefits external task environments.
By Linji Hao, Yanji He, Zihan Wang, Hansi Yang, Jiayuan Zhang, Hongyan Hao, Qi Gu, Yan Feng, Han Wu
The paper introduces AGORA, a framework that independently varies team composition and interdependence structures to study how social scaffolding influences the evolution of large language model agents. Experiments across web search, codebase understanding, and document QA show that appropriate scaffolding can outperform a strong agent evolved alone by up to 25 points, primarily by empowering the most capable member to lead while others step back. The findings suggest that designing a self‑evolving multi‑agent system requires careful consideration of its social scaffolding to shape learning dynamics and leadership emergence.
By Seojin Hwang, Youngmin Choi, Hwanhee Lee
Delta Tokenizer is a CPU‑side tokenization front end that only processes the difference between successive requests in a session, preserving the earlier token sequence and encoding only the appended text and its short tail. For long, stateless requests it splits the text into independent chunks, encodes them in parallel across multiple cores, and concatenates the results exactly. This approach reduces the time to first token by about 40× compared to the Hugging Face reference tokenizer, speeds up a four‑million‑character request 8.2× on 16 cores, and cuts median TTFT in SGLang chat by 59% while dramatically lowering P99 TTFT under load.
By Zhenyu Zhang, Zhichao Cao
EGCForest is a multi‑view hierarchical memory framework designed for enterprise group chats. It organizes the immutable message log into three complementary forests—Channel, Actor, and Artifact—each capturing temporal workflows, speaker‑conditioned context, and deliverable lifecycles, respectively. By using a Placement Map and sparse structural bridges, EGCForest jointly leverages message‑level relevance, forest summaries, and cross‑view links to retrieve and assemble distributed evidence, outperforming RAG and memory‑system baselines by up to 20.85 points on three benchmarks.
By Lingxi Cui, Jinhao Ruan, Wenqing Wei, Guo Yi, Li Rao, Ke Chen, Lidan Shou
The paper presents an industrial pipeline that converts U.S. state legislation into strict, schema‑constrained XML, emphasizing that the accuracy of the conversion stems more from the deterministic harness engineering than from the underlying language model. The authors conduct a controlled ablation study across three harnesses—zero‑shot, a single tool‑using agent, and a decomposed multi‑agent pipeline—paired with three Claude models on 153 gold bills from two jurisdictions. Results show that the full pipeline with the weakest model outperforms the strongest model alone (0.948 vs. 0.806), and that the harness narrows the model performance gap across four scored dimensions, with the remaining discrepancy largely due to citation normalization handled by an external lookup tool.
By Palvika Bansal, Rittika Jindal, Ruichong Wang
The paper introduces plug‑in decision heads that read hidden states from a frozen vision‑language model to predict web‑agent actions, bypassing the need to generate action‑specific text. These heads determine action type, target element, and text options, and a confidence cascade decides when to execute the head’s prediction versus letting the generator continue. Experiments with Qwen3‑VL show that the heads can decide 12–64 % of steps while keeping performance within 0.3–1.4 points of the full generator, and a looser threshold improves success on unseen sites and lifts WebArena task success with fewer generated tokens.
By Chengguang Gan, Yunhao Liang, Hanjun Wei, Qinghao Zhang, Zhixi Cai, Shiwen Ni
The paper introduces HICO-Bench, a benchmark of 180 everyday coordination scenarios where two agents with private information converse and then record a shared plan. A program verifies that the plans are feasible and an LLM judge checks that the conversation supports the plan, defining a Verified Success. Using Paired-Credit Policy Optimization (PaCPO), the authors improve the Verified Success rate of Qwen3-8B from 20.4% to 28.3%.
By Caixin Kang, Ke Wang, Yuchuan Wu, Xiaoqiang Wang, Ling Li, Bingxue Qiu, Yongbin Li, Yoichi Sato
The paper investigates general‑purpose coding agents that generate images by writing and executing programs, contrasting them with conventional generative models. On the AgentDrawBench benchmark, these agents excel at preserving structure and meeting precise visual constraints, achieving competitive performance in tasks where compositional accuracy outweighs photorealism, though they fall short in perceptual richness. The study further shows that the agents’ visual code can serve as an external language of thought, retaining substantial information for visual question answering and suggesting an internal, compositionally structured representation that guides visual operations across different drawing interfaces.
By Hengguang Zhou, Yuanhao Ban, Jiaqi Feng, Xiaohuan Pei, Justin Cui, Cho-Jui Hsieh
The paper argues that accuracy is an inadequate metric for certifying a coordinator’s theory‑of‑mind in multi‑agent systems, because the required accuracy depends on deployment‑specific costs and errors are often dependent. Experiments on task assignment, Level‑Based Foraging, Hanabi, and human judgments show that accuracy can both approve harmful language‑model deployments and reject beneficial ones. The authors propose a Blind‑Spot Gate that evaluates error costs, pairwise interactions, and dependence without deployment, achieving better safety outcomes than accuracy alone.
By Will Beaumaster, Paul Schrater
arXiv:2610.20365v1 Announce Type: new
Abstract: AI agent communities sustain lexical persistence, popular terms spread, recur, and achieve broad adoption, but whether they also develop the temporal s...
By Chih-Han Lin, Yu-Yun Chang, Po-Ya Angela Wang, Shu-Kai Hsieh
arXiv:2610.20474v1 Announce Type: new
Abstract: Pretrained vision-language-action models (VLAs) often fail midway through unseen long-horizon tasks. Recursive self-improvement (RSI) offers a remedy,...
By Peidong Wang, Ying Chang, Zhiming Ma, Xiaocui Yang, Shi Feng, Daqi Liu, Yuzhe Yang, Xu Cao, Chuang Wang, Jian Su, Daling Wang
arXiv:2610.19108v1 Announce Type: cross
Abstract: Multimodal Large Language Models (MLLMs) have been widely applied in embodied agents and robotics for vision tasks. However, MLLMs suffer from limita...
By Mahdieh Mallahnezhad, Masoud Jafaripour, Mohamadali Tofigh, Sheila Schoepp, Osmar R. Zaiane
arXiv:2610.19475v1 Announce Type: cross
Abstract: Multimodal in-context learning (ICL) is an important capability of multimodal large language models (MLLMs). Its performance is closely tied to the d...
By Yanshu Li, Ruixiang Tang, Junyi Jessy Li, Elias Stengel-Eskin
arXiv:2610.20563v1 Announce Type: cross
Abstract: Automated research with LLM agents has shown promising results in various domains. Through iterative refinement and scored verifiers, LLM agents are...
By Shangyin Tan, Ben Zhou, Lakshya A Agrawal, Jialin Zhang, Yilin Chen, Qiuyang Mang, Donghyun Lee, Dennis Jacob, Koushik Sen, Alex Dimakis, Matei Zaharia
arXiv:2601.10085v3 Announce Type: replace
Abstract: Conversational dynamics describe how rapport, emotion, and goals change over a dialogue. These changes are important in therapeutic conversations,...
By Viet Cuong Nguyen, Nhi Yen Nguyen, Kristin A. Candan, Mary Conlon, Vanessa Rumie, Kristen Risola, Michael L. Birnbaum, Munmun De Choudhury
arXiv:2604.23069v2 Announce Type: replace
Abstract: Large language model (LLM) agents often struggle in long-context interactions. As the history grows, sliding windows and prompt compression may dro...
By Yating Wu, Yuhao Zhang, Sayan Ghosh, Sourya Basu, Anoop Deoras, Jun Huan, Gaurav Gupta