https://radar.yuedu.biz本期有 7 条生产或规模化信号达到 A/B 级。高价值信息的共同点，是同时披露业务场景、运行机制与可复核结果。https://arxiv.org/html/2608.10224https://arxiv.org/html/2608.10224Wed, 12 Aug 2026 00:00:00 GMTLinkedIn 在真实企业客服流量上部署可持续演进的 Agent，把提示词、检索与评估做成可版本化闭环。两周随机 A/B 测试显示，QA 自助率提升 9.0 个百分点、取消流程自助率提升 4.8 个百分点、路由准确率提升 30.6 个百分点，并包含分阶段发布、回归检查与回滚机制。https://www.tezign.com/industries/beauty-brand-gea-sku-productionhttps://www.tezign.com/industries/beauty-brand-gea-sku-productionWed, 12 Aug 2026 00:00:00 GMT某美妆品牌将 GEA 企业智能体投入 80 个 SKU 的详情页生产工作流，制作周期从预计 16 周缩短至 3 周，A/B 测试变体由平均 1 个增加到 4 个；案例来自供应商官方页面，客户匿名，结果仍需独立核验。https://arxiv.org/abs/2608.24040https://arxiv.org/abs/2608.24040Tue, 25 Aug 2026 03:57:56 GMTEnterprise AI agents in production often need to be bounded, stateful, observable, and governable rather than fully autonomous. We present PinSieve, a production case study in a large-scale content-quality pipeline. Its…https://arxiv.org/abs/2608.23992https://arxiv.org/abs/2608.23992Tue, 25 Aug 2026 02:33:44 GMTLarge language model (LLM) agents invoke external tools to retrieve and reason over information beyond pretrained knowledge. The Model Context Protocol (MCP) standardizes how such tools are surfaced, and a proxy MCP ser…https://arxiv.org/pdf/2606.26859https://arxiv.org/pdf/2606.26859Fri, 26 Jun 2026 00:00:00 GMT快手把 AI AgentX 生产部署到主站推荐和本地生活业务三周，三个 Agent worker 将 374 个想法收敛成 10 次可上线实验，报告 8 倍并发、相对人工工程师 3.7 倍业务价值、0.561% 用户时长提升，以及超过人民币 1 亿元年化收入。论文没有提供可比基线。https://arxiv.org/abs/2608.26882https://arxiv.org/abs/2608.26882Thu, 27 Aug 2026 09:40:11 GMTIndustrial control systems (ICSs) rely on programmable logic controllers (PLCs) to connect networked computation with physical control. Tool-using large language model (LLM) agents represent an emerging attack threat: c…https://arxiv.org/abs/2608.24218https://arxiv.org/abs/2608.24218Tue, 25 Aug 2026 08:23:45 GMTEnterprise entity alignment must handle semi-structured records, implicit attributes, and unit or granularity mismatches. Manual matching is still common in practice, but does not scale as schemas and providers evolve…https://www.salesforce.com/ap/news/press-releases/2026/08/27/salesforce-and-anthropic-announce-claudeforce-the-1-ai-meets-the-1-ai-crmhttps://www.salesforce.com/ap/news/press-releases/2026/08/27/salesforce-and-anthropic-announce-claudeforce-the-1-ai-meets-the-1-ai-crmThu, 27 Aug 2026 00:00:00 GMTSalesforce 与 Anthropic 将企业 AI Claude 接入 Salesforce 数据、业务规则和工作流，并以 37 项销售技能开始试点。Salesforce 同时披露内部生产部署的 Slackbot 带来 810 万小时年化生产力收益；外部客户效果尚未披露，因此属于产品发布与内部采用证据。https://arxiv.org/html/2608.11241https://arxiv.org/html/2608.11241Sat, 15 Aug 2026 00:00:00 GMT系统在腾讯三个推荐业务线生产部署 78 天，记录 1,624 次工具调用。两个业务线的新业务接入时间由约 14 天缩短至约 3 天；论文明确说明该数字来自作者回忆、没有受控前序基线，应作为方向性案例而非普遍结论。https://doi.org/10.48550/arxiv.2606.28070https://doi.org/10.48550/arxiv.2606.28070Fri, 26 Jun 2026 00:00:00 GMT京东在电商核心业务的数百亿 SKU 上生产部署 LLM/VLM 商品知识平台，每天处理数亿次商品更新。论文披露知识生产精确率 94.2%、召回率 82.8%，商品信息质量问题下降 37%，核心属性自动填充率超过 80%，商品创意优化点击率提升约 9%。https://arxiv.org/abs/2608.22631https://arxiv.org/abs/2608.22631Wed, 26 Aug 2026 21:18:13 GMTTerminal agents are a compelling application of large language models (LLMs), with the potential to integrate deeply into users' daily workflows. Reinforcement learning (RL) is a key technique for improving their capabi…https://arxiv.org/abs/2608.24764https://arxiv.org/abs/2608.24764Tue, 25 Aug 2026 16:03:02 GMTLarge language model agents are moving beyond conventional retrieval-augmented generation toward direct interaction with external corpora. Direct Corpus Interaction (DCI) keeps the full corpus accessible, yet reachable…https://arxiv.org/abs/2608.22529https://arxiv.org/abs/2608.22529Sun, 23 Aug 2026 17:59:19 GMTEvaluating Large Language Model (LLM) code generation quality requires examining not just whether the generated code is correct, but whether it is maintainable, efficient, and stylistically sound, all of which are quali…https://arxiv.org/abs/2608.26885https://arxiv.org/abs/2608.26885Thu, 27 Aug 2026 09:42:39 GMTBackground. Large language models (LLMs) are increasingly used for screening in evidence synthesis, where false negatives can remove relevant studies before full-text assessment. We compared human and LLM title-and-abst…https://arxiv.org/abs/2608.25920https://arxiv.org/abs/2608.25920Wed, 26 Aug 2026 15:33:47 GMTAs large language model (LLM)-based multi-agent systems (MASs) are increasingly applied to long-horizon complex tasks, their reliability has emerged as the core bottleneck hindering their real-world deployment. Existing…https://arxiv.org/abs/2608.24569https://arxiv.org/abs/2608.24569Tue, 25 Aug 2026 13:51:52 GMTLarge language model (LLM) agents coordinate complex tasks through multi-role and multi-stage workflows. Upstream state is repeatedly transformed into intermediate language artifacts, such as summaries, plans, tickets…https://arxiv.org/abs/2608.23200https://arxiv.org/abs/2608.23200Tue, 25 Aug 2026 12:35:20 GMTLarge language models are increasingly expected to execute complex workflows whose success depends on maintaining interdependent constraints and producing artifacts that satisfy strict end-to-end verification. Yet succe…https://arxiv.org/abs/2608.23270https://arxiv.org/abs/2608.23270Mon, 24 Aug 2026 13:57:30 GMTWe present a new Astro-COLIBRI component that converts free-text GCN Circulars into structured, event-linked follow-up records and combines them with structured reports submitted directly by the community. A continuousl…https://arxiv.org/abs/2608.25602https://arxiv.org/abs/2608.25602Wed, 26 Aug 2026 10:23:02 GMTGenerative AI relocates the fixed cost of automation. A model provider pays to train a frontier system, while a downstream firm rents capability by usage; the same firm must carry a continuing payroll to supply a human-…https://arxiv.org/abs/2608.22068https://arxiv.org/abs/2608.22068Sat, 22 Aug 2026 18:18:07 GMTGeothermal well arrays, which organize multiple geothermal wells into carefully planned geometric configurations, provide opportunities to enhance energy production capacity and increase fault tolerance. The development…https://arxiv.org/abs/2608.27165https://arxiv.org/abs/2608.27165Thu, 27 Aug 2026 14:17:14 GMTAutoregressive large language models (LLMs) routinely generate factually incorrect outputs with high decoding confidence, limiting their deployment in high-stakes workflows. Existing output-stage uncertainty metrics can…https://arxiv.org/abs/2608.26417https://arxiv.org/abs/2608.26417Wed, 26 Aug 2026 21:35:14 GMTMetasurface design increasingly requires fast models that can operate across structurally distinct device families, rather than retraining a separate surrogate for every geometry class. Conventional neural network surro…https://arxiv.org/abs/2608.25869https://arxiv.org/abs/2608.25869Wed, 26 Aug 2026 14:41:24 GMTLarge language models (LLMs) increasingly assess generated content, giving rise to the LLM-as-a-Judge paradigm. These systems now score outputs, filter content, and gate iterative refinement in production pipelines, whe…https://arxiv.org/abs/2608.25697https://arxiv.org/abs/2608.25697Wed, 26 Aug 2026 12:13:05 GMTLarge language models (LLMs) are increasingly deployed with layered defenses, yet malicious prompts can still bypass them. Interpretability methods can expose model-internal signals along the generation path that could…https://arxiv.org/abs/2608.25523https://arxiv.org/abs/2608.25523Wed, 26 Aug 2026 08:33:10 GMTPrefix caching introduces a fundamental tradeoff in multi-agent large language model (LLM) serving: retaining a long system-prompt key-value (KV) cache for an agent accelerates future calls, yet it reduces the GPU memor…https://arxiv.org/abs/2608.24976https://arxiv.org/abs/2608.24976Tue, 25 Aug 2026 16:03:01 GMTPurpose: The digitization of historical documents presents fundamental challenges for modern information retrieval and Artificial Intelligence (AI) systems. Optical character recognition (OCR) errors in source corpora p…https://arxiv.org/abs/2608.24215https://arxiv.org/abs/2608.24215Tue, 25 Aug 2026 08:21:58 GMTLarge language model (LLM)-based multi-agent social simulation has demonstrated compelling results, but Agentopia was evaluated with 100 agents over 10 simulated years using Qwen3.5-397B-A17B, leaving the behavior of re…https://arxiv.org/abs/2608.23763https://arxiv.org/abs/2608.23763Mon, 24 Aug 2026 18:54:42 GMTThe Model Context Protocol (MCP) has emerged as the standard layer connecting Large Language Model agents to external tool backends. This openness introduces a severe server-side threat we term TrustShift: a compromised…https://arxiv.org/abs/2608.23653https://arxiv.org/abs/2608.23653Mon, 24 Aug 2026 11:50:07 GMTAI agents are increasingly used for simulation-driven engineering. Physical system modeling presents different requirements from general-purpose code generation in software engineering, because correctness depends not o…https://arxiv.org/abs/2608.23058https://arxiv.org/abs/2608.23058Mon, 24 Aug 2026 10:02:39 GMTLarge language models (LLMs) now support forecasting systems that combine language-based reasoning with temporal data, evidence retrieval, external tools, and iterative prediction. We investigate LLM-based forecasting a…