技術深讀
RAG、AI Agent、碳排放等技術深入探討
共 107 篇文章
Agent 記憶中毒的成本量化與內容篩選的極限
# Agent 記憶中毒的成本量化與內容篩選的極限 **來源：** arXiv（cs.CR, cs.AI） **日期：** 2026-08-21 ## 摘要 這篇論文量化了 Agent 持久記憶的「記憶中毒」攻擊成本：錯誤陳述一旦被寫入長期記憶，會在未來相關會話中被反覆檢索，造成錯誤資訊的持久化。只用一次性生成的普通假陳述污染 LongMemEval 語料 1.2%，準確率就從 0.850 ...
AI with Authority：形式驗證從成本變生產力必需品
# AI with Authority：形式驗證從成本變生產力必需品 **來源：** arXiv（cs.SE, cs.AI, cs.AR, cs.LO） **日期：** 2026-08-21 ## 摘要 六十年來機器驗證一直是巨大成本開銷，只負擔得起極少數「例外工件」。這篇報告指出生成式 AI 反轉了此關係：在 AI 速度下，機器驗證不只經濟、更是生產力的必需品——它是讓「一個人安全地大規模...
AI-to-AI Code Review：誰來審查審查者
# AI-to-AI Code Review：誰來審查審查者 **來源：** arXiv（cs.SE） **日期：** 2026-08-21 ## 摘要 AI 編碼代理已整合進軟體開發工作流，並在 PR 兩端同時運作：AI 作者建立/修改 PR、AI 審查者評估——形成「一個 AI 代理審查另一個 AI 代理貢獻」的封閉迴圈。研究連結 CodAGE 資料集中 AI 歸屬的 PR 與審查事件，...
AID-Guard：把「一次批准」鎖定為「最多一個效應」的狀態式授權
# AID-Guard：把「一次批准」鎖定為「最多一個效應」的狀態式授權 **來源：** arXiv（cs.CR, cs.AI） **日期：** 2026-08-21 ## 摘要 使用工具的 AI 代理把委派任務轉成「供應商效應」，但授權往往在「准入」階段就結束，而供應商狀態、交付、重試、復原仍在演化——請求可能在提交前改變，或回應遺失導致一次批准被執行兩次（重複扣款/重複下單）。AID-G...
Artic：把自然語言工作流編譯成「工件驅動」以可靠執行
# Artic：把自然語言工作流編譯成「工件驅動」以可靠執行 **來源：** arXiv（cs.SE） **日期：** 2026-08-21 ## 摘要 自然語言工作流給代理「軟體式介面」，但承諾尚不可靠：工作流描述常把資料依賴隱含化，執行者必須自行推斷每一步該用哪個先前結果；代理也會在上下文壓力下無法遵循長或分支指令。Artic 是「工件驅動」工作流編譯器：把自然語言工作流轉成每個步驟明確...
圖工程（Graph Engineering）：從個體智能到系統智能的代理範式
# 圖工程（Graph Engineering）：從個體智能到系統智能的代理範式 **來源：** arXiv（cs.IR, cs.AI, cs.ET） **日期：** 2026-08-21 ## 摘要 LLM 已從語言生成器演進成能執行複雜長程任務的自主代理，陸續誕生 Prompt Engineering（激發能力）、Context Engineering（管理資訊存取）、Harness E...
看不見的代理、不知情的病患：負責任 AI 部署的三個最低標準
# 看不見的代理、不知情的病患：負責任 AI 部署的三個最低標準 **來源：** arXiv（cs.CY） **日期：** 2026-08-21 ## 摘要 自主 AI 診斷代理（分析病患臨床資料並在無強制即時人工審查下輸出診斷或分流決策）正以超越治理基礎設施的速度在撒哈拉以南非洲的 eHealth 平台部署。現有 AI 問責、透明與可解釋框架多以臨床醫師為中心，並假設低資源情境不存在的監管...
規格可攜性：跨 LLM 開發代理的規格不是「代理中立」工件
# 規格可攜性：跨 LLM 開發代理的規格不是「代理中立」工件 **來源：** arXiv（cs.SE, cs.AI, cs.LO） **日期：** 2026-08-21 ## 摘要 以 Oracle→PostgreSQL 遷移為受控軟體轉型任務，研究跨代理的「規格可攜性」。第一階段「規格優先」遷移管線在 1,006 個 PL/SQL 檔案上，623 個成功重新生成、380 個產出腳本在 P...
研究觀察-2026-08-22：七日 digest 跨領域趨勢回顧
# 研究觀察 2026-08-22 — 七日知識掃描（08/16–08/22）跨領域趨勢 > 回顧 7 份 digest（2026-08-16 ~ 08-22），整理三條跨領域主線：AI 成本紀律 × 計費基礎設施、能源 × AI×ESG 工程化、AI 採用率敘事校正 × 人才組織工程。純 digest 回顧，未另行搜尋網路。與 08-19 觀察相較，三條主線均有新數據加厚。 ## 趨勢一：A...
Agent 治理框架雙軌計畫（內部紀律 × 對外產品）
# Agent 治理框架雙軌計畫 > 一句話：把「我們如何管理自己的 31 個自動化 agent」變成可重現的治理框架（內部紀律），同時包裝成 Pilot Trap 診斷產品（對外營收）。源自 2026-08-19 麥肯錫《逃離 Pilot 陷阱》報告分析。 ## 核心洞察（差異化主張） 台灣多數企業正卡在麥肯錫說的「50 個 pilot」階段——agent 遍地開花、無人治理、舊流程碎片被...
Agent 事故教訓庫（Incident Log）
# Agent 事故教訓庫 > W2-A4 交付物。真實踩過的坑，格式：**症狀 → 根因 → 修復 → 預防檢查項**。用途：(1) 月檢的檢查清單來源 (2) 對外「我們怎麼管自己的 agent」的實戰素材 (3) 新 agent 上線前的風險對照表。 ## 1. Skill 自動覆寫人工編輯 - **症狀**：auto-commit skill 把未完成的 PR 合進 main，破壞主...
Agent 資產盤點表（2026-08-20）
# Agent 資產盤點表 > W1-A1 交付物。盤點範圍：RongRise 營運的 Hermes Agent 全資產。目的：讓每個自動化資產有 owner、可審計、可評估、可退役——為「治理五檢查」提供基準線。 ## 一、資產總覽（2026-08-20 實測） | 資產類型 | 數量 | 備註 | |---------|------|------| | Cron jobs | **27*...
Agent 資產月檢報告（2026-08）
# Agent 資產月檢報告 — 2026-08 > 執行者：[bot:ops-watch] Agent 資產月檢（49a80c05d036，首次執行）。依據治理五檢查（鋪路 or 硬化 / owner / 可審計 / 划算 / 退役）逐一過關。 ## 一、總覽 | 項目 | 數值 | |------|------| | Cron jobs 總數 | **33**（enabled=27 / ...
Agentic HR 起步診斷工作坊 — 交付包（B2）
# Agentic HR 起步診斷工作坊 — 交付包 > W4-B2 交付物。把「Pilot Trap 診斷量表」升級成可收費的顧問服務。 > 定位差異：**不是課程，是診斷 + 落地藍圖**（閃開天下學習的「虛擬員工課程」對打）。 ## 產品定位 - **一句話**：三小時，把你的 AI 部隊雷達圖變成「AI 轉型 North Star 一頁圖」+ 30/60/90 天行動計畫。 - **...
研究觀察-2026-08-19：七日 digest 跨領域趨勢回顧
# 研究觀察 2026-08-19 — 七日知識掃描（08/13–08/19）跨領域趨勢 > 回顧 7 份 digest（2026-08-13 ~ 08-19），整理三條跨領域主線：AI 成本重估 × 財務管理、AI 擴張 × 能源 ESG、AI 時代 × 人才重新定價。純 digest 回顧，未另行搜尋網路。 ## 趨勢一：AI 從「買算力」轉向「管成本」— 企業 AI 經濟學成形 **現...
A Graph-Based Reinforcement Learning Framework for Structured Drift Diagnosis and Recovery in Autonomous LLM Agents
# A Graph-Based RL Framework for Structured Drift Diagnosis and Recovery in Autonomous LLM Agents **作者：** Ismail El Hamraoui, Sagar Jose, Nicolas Bureau, Robert Plana **發表：** 14 Aug 2026 | **引用：** N/...
AI Adoption in S&P 500 Firms
# AI Adoption in S&P 500 Firms **作者：** Yang Yu, Martin Fleming, Lucy Hampton, Christophe Combemale, Neil Thompson **發表：** 09 Jul 2026 | **引用：** N/A（Semantic Scholar 限流未取得） ## 摘要 本研究以 SEC 10-K 申報（法律禁...
AI-Augmented Human Resource Management? Insights from German companies
# AI-Augmented Human Resource Management? Insights from German companies **作者：** Yannick Kalff, Katharina Simbeck **發表：** 15 Jul 2026 | **引用：** N/A（Semantic Scholar 未索引） ## 摘要 本研究以訪談、焦點團體與問卷（N=410）探...
AI-Increased Talent Retention Strategies: Fostering Long-Term Employee Engagement and Development in Talent Management
# AI-Increased Talent Retention Strategies **作者：** Jay Barach **發表：** 22 Jul 2026 | **引用：** N/A（Semantic Scholar 未索引） ## 摘要 本文以系統性文獻回顧梳理 AI 在人才留任（Talent Retention）的整合：預測模型、情緒分析、個人化職涯規劃等工具可預測員工離職傾向與最...
The CASE Framework: A Multi-Disciplinary Control Architecture for Governing Enterprise Agentic AI
# The CASE Framework: A Multi-Disciplinary Control Architecture for Governing Enterprise Agentic AI **作者：** Srinivas Telukunta, Georgios Nektarios Lilis, Lucio Baron **發表：** 10 Aug 2026 | **引用：** N/A...
Chitchat with AI: Understand the supply chain carbon disclosure of companies worldwide through Large Language Model
# Chitchat with AI: Supply Chain Carbon Disclosure Analysis with LLM **作者：** Haotian Hang, Yueyang Shen, Vicky Zhu, Jose Cruz, Michelle Li **發表：** 26 Oct 2025 | **引用：** N/A（Semantic Scholar 限流未取得） #...
The Deployment Wall: A Diagnostic Framework and Instrument for Enterprise AI in the Deployment Era
# The Deployment Wall: A Diagnostic Framework and Instrument for Enterprise AI in the Deployment Era **作者：** Fabricio F. Costa **發表：** 31 Jul 2026 | **引用：** N/A（新論文，Semantic Scholar 未索引） ## 摘要 企業 Ge...
EmeraldMind: A Knowledge Graph-Augmented Framework for Greenwashing Detection
# EmeraldMind: A Knowledge Graph-Augmented Framework for Greenwashing Detection **作者：** Georgios Kaoukis, Ioannis Aris Koufopoulos, Eleni Psaroudaki, Danae Pla Karidi, Evaggelia Pitoura, George Papas...
A Gateway Architecture for Enterprise MCP Authentication: Unifying Heterogeneous Auth, Identity Delegation, and the User / Non-User Persona Problem
# A Gateway Architecture for Enterprise MCP Authentication **作者：** Suraj Kumar, Amy Wang, Srinivasan Manoharan **發表：** 11 Aug 2026 | **引用：** N/A（新論文） ## 摘要 MCP 已成為企業 LLM 代理連接工具的預設介面，一年內大型組織從零暴增到數十個內...
ESG Beliefs of Large Language Models: Evidence and Impact
# ESG Beliefs of Large Language Models: Evidence and Impact **作者：** Tong Li, Luping Yu **發表：** 26 Dec 2025 | **引用：** N/A（新論文） ## 摘要 研究 LLM 是否對 ESG 議題持有系統性信念，以及這些信念如何與人類市場參與者比較、甚至反過來影響人類。基於原本施測於專業與散戶...
Green SARC: Predictive Cost and Carbon Governance for Agentic AI Systems
# Green SARC: Predictive Cost and Carbon Governance for Agentic AI Systems **作者：** Gaston Besanson **發表：** 14 Jun 2026 | **引用：** N/A（新論文） ## 摘要 Agentic AI 系統透過工具與子代理行動，但控制其財務與環境成本的機制仍停在「執行後才看的儀表板」。G...
How Organizations Use AI: Evidence from ChatGPT
# How Organizations Use AI: Evidence from ChatGPT **作者：** Aaron Chatterji, David Holtz, Neel Rakholia, Prasanna Tambe, Gawesha Weeratunga **發表：** 12 Aug 2026 | **引用：** N/A（新論文，Semantic Scholar 未索引） ...
Human-AI Technology Integration and Green ESG Performance: Evidence from Chinese Retail Enterprises
# Human-AI Technology Integration and Green ESG Performance **作者：** Jun Cui **發表：** 03 Jul 2025 | **引用：** N/A（Semantic Scholar 限流未取得） ## 摘要 本研究以 2019–2023 年 5,400 筆公司-年度面板資料（CNRDS/CSMAR），用固定效應迴歸檢驗中國...
InFactPlanner: Planning Sustainable Geo-Distributed LLM Data Centers
# InFactPlanner: Planning Sustainable Geo-Distributed LLM Data Centers **作者：** Nicoletta Tsiopani, Moysis Symeonides, George Pallis, Marios D. Dikaiakos **發表：** 13 Aug 2026 | **引用：** 0（Semantic Schol...
Self-evolving Agentic Customer Support System at LinkedIn
# Self-evolving Agentic Customer Support System at LinkedIn **作者：** Chih Hui Wang, Mengdie Tu, Qianyun Zhang, Wei Wu, Lili Zhou, Mingqi Shen, Changshuai Wei **發表：** 10 Aug 2026 | **引用：** N/A（新論文） ##...
Mandato: Protocol-Level Enforcement of Digitally Signed Mandates on AI Agent Actions with Cryptographically Chained Audit Trails
# Mandato: Protocol-Level Enforcement of Digitally Signed Mandates on AI Agent Actions **作者：** Giovanni Racioppi **發表：** 14 Aug 2026 | **引用：** N/A（新論文） ## 摘要 AI 代理透過 MCP 等標準工具呼叫協定操作外部系統，但授權邏輯散落在應用程式...
Polaris: Multi Agentic System for Conversational Enterprise Analytics
# Polaris: Multi Agentic System for Conversational Enterprise Analytics **作者：** Varuni H K, Soham Sarkar, Jay Kumar, Goutham Krishnan, Tanvi Johari, Avinash Bharadwaj, Santosh Hegde **發表：** 14 Aug 20...
Upskilling with Generative AI: Practices and Challenges for Freelance Knowledge Workers
# Upskilling with Generative AI: Practices and Challenges for Freelance Knowledge Workers **作者：** Kashif Imteyaz, Isabel Lopez, Nakul Rajpal, Hunjun Shin, Saiph Savage **發表：** 29 Apr 2026 | **引用：** 0...
AI 記憶管理全景：研究與實務技巧
# AI 記憶管理全景：研究與實務技巧（2026-08） > 來源：彙整 2026 年 AI agent memory 研究 Survey、8 大記憶框架比較、Context Engineering 官方指南與社群討論共 14 篇（來源清單見文末） > 存入：2026-08-11｜分類：技術深讀 ## 一句話 AI 的記憶能力，是 2026 年「聊天機器人」與「數位員工」之間的分水嶺——沒有...
An End-to-End Agent Auditing Engine (A²E)
# An End-to-End Agent Auditing Engine **作者：** Haoning Wang, Mingxun Zhang, Chenyue Yu, Yingjun Shang, Xia Hu, Guanchu Wang, Na Zou **發表：** 07 Aug 2026 | **引用：** N/A（新論文，Semantic Scholar 尚未索引） ## 摘要 ...
Agentic AI Sustainability Assessment for Supply Chain Document Insights
# Agentic AI Sustainability Assessment for Supply Chain Document Insights **作者：** Diego Gosmar, Anna Chiara Pallotta, Giovanni Zenezini **發表：** 10 Nov 2025 | **引用：** 2 ## 摘要 本文提出以 Agentic AI 為核心的供應鏈...
Algorithmic Criminal Liability in Greenwashing: Comparing India, United States, and European Union
# Algorithmic Criminal Liability in Greenwashing: Comparing India, United States, and European Union **作者：** Sahibpreet Singh, Manjit Singh **發表：** 14 Dec 2025 | **引用：** 1 ## 摘要 AI 驅動的漂綠（greenwashin...
Estimating GHG Emissions from AI Use: Framework for Corporate-Level Measurement
# Estimating GHG Emissions from AI Use: Framework for Corporate-Level Measurement **作者：** John Bistline, Shaena Ulissi, Steven J. Davis, Jonathan Glidden, James Joyce, Mo Li, Jackson Mohsenin, Sangwo...
Evaluating LLM Trade-offs for Enterprise Automation: Lessons from Workflow Generation in a Production Enterprise Platform
# Evaluating LLM Trade-offs for Enterprise Automation: Lessons from Workflow Generation in a Production Enterprise Platform **作者：** Xavier Wrenn, Radoslav Raykov, Aleksandar Angelov, Hirokuni Kitahar...
A Security-Oriented Lifecycle Model for Large Language Model Systems
# A Security-Oriented Lifecycle Model for Large Language Model Systems **作者：** Eleftherios Batzolis, George Drosatos, Vassilis Katsouros, Konstantinos Rantos **發表：** 04 Aug 2026 | **引用：** 0 ## 摘要 LL...
Scoping Review of AI, Metrology, and ESG in the Semiconductor Sector: Implications for Safe and Sustainable by Design (SSbD)
# Scoping Review of AI, Metrology, and ESG in the Semiconductor Sector: Implications for Safe and Sustainable by Design (SSbD) **作者：** Karen Ang, Han-Teng Liao **發表：** 25 Jul 2026 | **引用：** N/A（新論文，S...
SkillProx: Self-Evolving Agent Skills via Proximal Textual Gradient Descent
# SkillProx: Self-Evolving Agent Skills via Proximal Textual Gradient Descent **作者：** Mingxuan Zheng, Yujin Zhou, Chuxue Cao, Boqin Yin, Yuyao Zhang, Jiapeng Sun, Shuaishuai Gong, Sirui Han, Yike Guo...
TEPA: Revoking Stale Memories for Conflict-Robust Language Agents
# TEPA: Revoking Stale Memories for Conflict-Robust Language Agents **作者：** Yan Zhou, Yue Ouyang, Kaiyang Zheng, Suncheng Xiang **發表：** 07 Aug 2026 | **引用：** N/A（新論文，Semantic Scholar 尚未索引） ## 摘要 長期記...
From Augmentation to Reconstruction: Guiding the AI Disruption to the Good Place
# From Augmentation to Reconstruction: Guiding the AI Disruption to the Good Place **作者：** David M. Rothschild, Jake M. Hofman, Markus Mobius, Brendan Lucier, Eleanor Dillon, Daniel G. Goldstein, Nic...
From Business Problems to AI Solutions: Where Does Transformation Support Fail
# From Business Problems to AI Solutions: Where Does Transformation Support Fail **作者：** Abir Trabelsi, Imen Benzarti, Hafedh Mili, Darine Ameyed **發表：** 20 Apr 2026 ## 摘要 把業務問題轉譯成規格良好的機器學習解決方案是 AI ...
Supporting Autonomous Process Execution within a Multi-Perspective Constraint Frame via Numeric Planning
# Supporting Autonomous Process Execution within a Multi-Perspective Constraint Frame via Numeric Planning **作者：** Paul Wittlinger, Giacomo Acitelli, Anti Alman, Fabrizio Maria Maggi, Andrea Marrella...
AI Adoption Across a Multinational Workforce: Sociotechnical Conditions for GenAI Acceptance in Human Resources
# AI Adoption Across a Multinational Workforce: Sociotechnical Conditions for GenAI Acceptance in Human Resources **作者：** Dalia Ali, Maria José Rodríguez Velázquez, Manoel Horta Ribeiro, Vera Liao, O...
Memory Provenance Laundering in LLM Agents: A Non-Amplification Firewall for Persistent Memory
# Memory Provenance Laundering in LLM Agents: A Non-Amplification Firewall for Persistent Memory **作者：** Jinghan Xu, Yiyong Xiao, Wanru Shao, Hankai Liu, Xinjin Li **發表：** 31 Jul 2026 ## 摘要 長期記憶讓 LL...
Tool Specifications Matter: Uncovering and Mitigating Safety Risks in AI Agents
# Tool Specifications Matter: Uncovering and Mitigating Safety Risks in AI Agents **作者：** Minghui Pan, Jiayuxuan Yang, Yuanyuan Yuan, Yu Jiang, Zhenpeng Chen **發表：** 31 Jul 2026 ## 摘要 LLM 一旦被部署為可呼叫工...
Beyond Component Testing: Validating Agentic AI Systems
# Beyond Component Testing: Validating Agentic AI Systems **作者：** Fabio Orazio Mirto, Luca D'Agati, Giuseppe Tricomi, Stefano Silvestri, Francesco Longo, Antonio Puliafito, Giovanni Merlino **發表：** 3...
Toward an AI-Powered Computational Testbed for Workforce Policy
# Toward an AI-Powered Computational Testbed for Workforce Policy **作者：** Sumer S. Vaid, Ashley V. Whillans **發表：** 18 May 2026 ## 摘要 勞動力轉型難以預測、管理不善的代價高昂。本文結合 LLM 生成式代理與管理科學、組織行為研究，提出「動態員工代理」：在同意前提下...
AI Agents for Sustainable SMEs: A Green ESG Assessment Framework
# AI Agents for Sustainable SMEs: A Green ESG Assessment Framework **作者：** Viet Trinh, Tan Nguyen, Minh-Huyen Phan, Quan Luu **發表：** 5 Apr 2026 ## 摘要 本研究提出以 AI 驅動的 ESG 評估框架，專為歐洲中小企業設計。第一階段從 Flash Eu...
A Research on Business Process Optimisation Model Integrating AI and Big Data Analytics
# A Research on Business Process Optimisation Model Integrating AI and Big Data Analytics **作者：** Di Liao, Ruijia Liang, Ziyi Ye **發表：** 12 Nov 2025 ## 摘要 本研究建構整合 AI 與大數據分析的流程優化模型，採用資料處理、AI 演算法、業務邏輯...
The impact of artificial intelligence on enterprise software user roles
# The impact of artificial intelligence on enterprise software user roles **作者：** Isabel Unger, Elizangela Valarini, Martin Schrepp, Nina Hollender, Gabriela Rocha, Erik Bertram **發表：** 24 Jun 2026 ...
Mapping the Landscape of AI-Driven Human Resource Management: A Social Network Analysis of Research Collaboration
# Mapping the Landscape of AI-Driven Human Resource Management **作者：** Mehrdad Maghsoudi, Motahareh Kamrani Shahri, Mehrdad Agha Mohammad Ali Kermani, Rahim Khanizad **發表：** 18 Aug 2023 ## 摘要 本研究運用社...
Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting
# Accounting for AI Inference in Corporate GHG Inventories **作者：** Guillermo Llopis (SOMA AI, Barcelona) **發表：** 9 Jun 2026 ## 摘要 AI 推論服務（API 訂閱、企業聊天工具、嵌入 AI 功能的 SaaS 產品）在 CSRD 規範下應屬於範疇三類別一，但目前並無標準核...
AI-Powered Sustainable Finance: An Integrative Taxonomy and Framework of AI Applications for Sustainable Investment Decisions
# AI-Powered Sustainable Finance: An Integrative Taxonomy and Framework **作者：** Eduardo C. Garrido-Merchán, Esther Vaquero Lafuente, Elisa Aracil **發表：** 25 May 2026 ## 摘要 本綜述提供 AI 在永續金融應用的完整分類法，涵蓋機...
ESG Reporting Lifecycle Management with Large Language Models and AI Agents
# ESG Reporting Lifecycle Management with Large Language Models and AI Agents **作者：** Thong Hoang, Mykhailo Klymenko, Xiwei Xu, Shidong Pan, Yi Ding, Xushuo Tang, Zhengyi Yang, Jieke Shi, David Lo **...
A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents
# A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents **作者：** Krti Tallam **發表：** 10 Jun 2026 ## 摘要 生產環境中的 AI Agent 挑戰了傳統資安邊界模型——Agent 讀取上下文、呼叫工具、連接器並修改系統記錄，風險存在於工作流程內...
Toward Trustworthy Evaluation of Sustainability Rating Methodologies: A Human-AI Collaborative Framework for Benchmark Dataset
# Toward Trustworthy Evaluation of Sustainability Rating Methodologies **作者：** Xiaoran Cai, Wang Yang, Xiyu Ren, Chekun Law, Rohit Sharma, Peng Qi **發表：** 19 Feb 2026 ## 摘要 不同 ESG 評級機構對同一家公司的評分差異極大，...
Making Agent-Mediated Contributions Governable: A Project-Level Governance Manifest for Open-Source AI Collaboration
# Making Agent-Mediated Contributions Governable: A Project-Level Governance Manifest for Open-Source AI Collaboration **作者：** Jinjin Gao, Luyang Li, Shufen Guo, Ligang He, Xiaoning Sun **發表：** 2026-...
A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance
# A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governance **作者：** Andrea Ferrario **發表：** 2026-07-17 ## 摘要 AI 治理越來越需要判斷一個 AI 系統是否能隨時間保持足夠的可信賴度、觀察到的變化是否可容忍、以及這些判斷應如何以透明且可爭議的方式記錄...
Cheaper AI, More Informality? A Dual Labor Market Model for Developing Economies
# Cheaper AI, More Informality? A Dual Labor Market Model **作者：** Gabriel Montes-Rojas, Fernando Toledo, Juan Manuel Rodríguez Repeti **發表：** 2026-07-16 ## 摘要 本文研究 AI 價格下降對勞動市場結果的影響，特別關注 AI 是否會創造正式就...
Inference Economics of Enterprise Coding Agents: Cloud vs. On-Premise LLMs
# Inference Economics of Enterprise Coding Agents: Cloud vs. On-Premise LLMs **作者：** Sheng-Wei Peng, Yi-Hsun Lin, Yi-Pei Lee **發表：** 2026-07-13 ## 摘要 自主程式碼代理讓工程組織面臨兩難選擇：API-based 的 Frontier 模型（強推理能力...
Frontier AI Performance Across the Business Disciplines: A Case-Grounded Benchmark of Knowledge Work and Analytical Reasoning
# Frontier AI Performance Across the Business Disciplines **作者：** Ajay Patel, Kartik Hosanagar, Ramayya Krishnan, Chris Callison-Burch, Karim Lakhani, Mitch Weiss **發表：** 2026-07-17 ## 摘要 LLM 在傳統基準測...
When Do Multi-Agent Systems Help? An Information Bottleneck Perspective
# When Do Multi-Agent Systems Help? An Information Bottleneck Perspective **作者：** Wendi Yu, Lianhao Zhou, Xiangjue Dong, Sai Sudarshan Barath, Declan Staunton, Byung-Jun Yoon, Xiaoning Qian, James Ca...
Noise is Signal: Density-Based Outliers as Leading Indicators of Occupational Emergence in Labor Market Text
# Noise is Signal: Density-Based Outliers as Leading Indicators of Occupational Emergence **作者：** Shreyash Rawat **發表：** 2026-06-22 ## 摘要 標準 NLP 職業分群流程會將密度式方法標記為雜訊的 10-15% 職缺資料丟棄。本文主張這是個錯誤：在快速演變的領域中...
Oracle Agent Memory as an Enterprise Memory Substrate for Long-Horizon AI Agents
# Oracle Agent Memory as an Enterprise Memory Substrate for Long-Horizon AI Agents **作者：** Richmond Alake, Cesare Bernardis, Paul Cayet, Luca Engel, Damien Hilloulin, Sungpack Hong, Allen Hosler, Nic...
QwenPaw-Data: Bridging Facts, Methodology, and Execution for Autonomous Enterprise Data Analytics
# QwenPaw-Data: Autonomous Enterprise Data Analytics **作者：** Tianjing Zeng, Yuntao Hong, Zhongjun Ding, Dandan Liu, Yinan Mei, Yunxiang Su, Yiming Wang, Xiaojian Zhang, Jingyu Zhu, Junhao Zhu, Zhuowe...
當 AI 開始做決策：企業如何避免「誰負責」變成無人回答的問題
> AI Agent 不是新員工，是一種新型風險載體。治理的起點，是先釐清它永遠不能承擔的那個責任。 ## 一、企業比自己以為的更沒準備好 Gartner 在 2025 年中做出預測：到 2027 年底，超過 40% 的代理式 AI（Agentic AI）專案會被砍掉——原因不是模型不好用，而是成本失控、商業價值說不清楚、風險控管沒跟上。多數代理式 AI 專案仍停在早期實驗或概念驗證階段，很大...
The impact of artificial intelligence on enterprise software user roles
# The impact of artificial intelligence on enterprise software user roles **作者：** Isabel Unger, Elizangela Valarini, Martin Schrepp, Nina Hollender, Gabriela Rocha, Erik Bertram **發表：** 2026-06-24 #...
Cheap Code, Costly Judgment: A Case Study on Governable Agentic Software Engineering
# Cheap Code, Costly Judgment: A Case Study on Governable Agentic Software Engineering **作者：** James C. Davis, Paschal C. Amusuo, Tanmay Singla, Berk Çakar, Kirsten A. Davis **發表：** 2026-07-01 ## 摘要...
Evolving the Productivity Equation: Should Digital Labor Be Considered a New Factor of Production?
# Evolving the Productivity Equation: Should Digital Labor Be Considered a New Factor of Production? **作者：** Alex Farach, Alexia Cambon, Jared Spataro **發表：** 2025-05-14 ## 摘要 傳統生產力指標難以捕捉 AI 的真實經濟影響...
The Iceberg Index: Measuring Skills-centered Exposure in the AI Economy
# The Iceberg Index: Measuring Skills-centered Exposure in the AI Economy **作者：** Ayush Chopra, Santanu Bhattacharya, DeAndrea Salvador, Ayan Paul, Teddy Wright, Aditi Garg, Feroz Ahmad, Alice C. Sch...
What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates
# What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates **作者：** Arman Ghaffarizadeh, Danyal Mohaddes, Aliakbar Izadkhah, Shahriar Noroozi...
The Organizational Behavior of Agentic AI: Collective Intelligence in Human-Agent Workflows
# The Organizational Behavior of Agentic AI: Collective Intelligence in Human-Agent Workflows **作者：** Canhui Liu **發表：** 2026-06-29 ## 摘要 Agentic AI 不再只是單一助手，而是以規劃者、解題者、審查者、記憶管理者、工具使用者、編排者等集體形式進入組織流...
Reasoning effort, not tool access, buys first-try reliability in agentic code generation
# Reasoning effort, not tool access, buys first-try reliability in agentic code generation **作者：** Achint Mehta **發表：** 2026-07-02 ## 摘要 這項實證研究測試了 Agentic coding assistant 是否真的因為更多工具（瀏覽器測試工具、設計提示）而產...
Steerability via constraints: a substrate for scalable oversight of coding agents
# Steerability via constraints: a substrate for scalable oversight of coding agents **作者：** Thomas Winninger **發表：** 2026-07-02 | **備註：** Accepted to ICML 2026 Deep Learning for Code Workshop ## 摘要 ...
The Urban-Rural Divide in the Age of AI: Assessing the Effects of Technology and Automation on Regional Labor Markets
# The Urban-Rural Divide in the Age of AI: Assessing the Effects of Technology and Automation on Regional Labor Markets **作者：** Chau Tran Bao, Khoi Nguyen Dinh Nguyen, Ha Nguyen Manh, Ngan Nguyen Thi...
Agent-Native Immune System: Architecture, Taxonomy, and Engineering
# Agent-Native Immune System: Architecture, Taxonomy, and Engineering **作者：** Bo Shen, Lifeng Chang, Tianyuan Wei, Yunpeng Li, Feng Shi, Yichen Han, Peijie Gao, Shiyi Kuang, Xin Chang, Dehui Li **發表：...
Govern the Repository, Not the Agent: Measuring Ecosystem-Level Risk in AI-Native Software
# Govern the Repository, Not the Agent: Measuring Ecosystem-Level Risk in AI-Native Software **作者：** Daniel Russo **發表：** 2026-06-26 ## 摘要 本文研究自主 coding agent 在共享儲存庫中開 PR 與 merge 時的生態系風險。透過分析 93 萬條 ...
The impact of artificial intelligence on enterprise software user roles
# The impact of artificial intelligence on enterprise software user roles **作者：** Isabel Unger, Elizangela Valarini, Martin Schrepp, Nina Hollender, Gabriela Rocha **發表：** 2026-06-24 ## 摘要 本研究以 SAP ...
Invoice Haystack: Benchmarking Document Retrieval and Visual Question Answering Under Strong Visual Homogeneity
# Invoice Haystack: Benchmarking Document Retrieval and Visual Question Answering Under Strong Visual Homogeneity **作者：** Heethanjan Kanagalingam, Thenukan Pathmanathan, Mokeeshan Vathanakumar, Basim...
From RAN Control to Agentic Intelligence: Architecture and Vision for Energy Efficient AI-RAN
# From RAN Control to Agentic Intelligence: Architecture and Vision for Energy Efficient AI-RAN **作者：** Sabrine Aroua, Alexis I. Aravanis, Ilias Chatzistefanidis, Hamza Abbar, Anh-Khoa Dang **發表：** 2...
Autonomous Event-Driven Multi-Agent Orchestration for Enterprise AI at Scale
# Autonomous Event-Driven Multi-Agent Orchestration for Enterprise AI at Scale **作者：** Harsh Rao Dhanyamraju, Leonidas Raghav, Aaron Lee **發表：** 2026-06-18 ## 摘要 本文評估了 DAG Plan-and-Execute 與 ReAct 兩...
AI4SE and SE4AI Exploration: A Decade Looking Back and Forward
# AI4SE and SE4AI Exploration: A Decade Looking Back and Forward **作者：** H. Sinan Bank, Daniel R. Herber, Thomas Bradley **發表：** 2026-06-17 ## 摘要 本文回顧 AI 與系統工程（SE）交會領域十年進展，劃分三個階段：基礎期、應用期、LLM 轉折期。202...
Deontic Policies for Runtime Governance of Agentic AI Systems
# Deontic Policies for Runtime Governance of Agentic AI Systems **作者：** Anupam Joshi, Tim Finin, Karuna Pande Joshi, Lalana Kagal **發表：** 2026-06-17 ## 摘要 本文針對 LLM 驅動的自主 Agent 系統提出 AgenticRei 治理框架，使...
IHBench: Evaluating Post-Interruption Recovery in Voice Agents with Structured Workflows
# IHBench: Evaluating Post-Interruption Recovery in Voice Agents with Structured Workflows **作者：** Ahmad Salimi, Wentao Ma, Yuzhi Tang, Dongming Shen, Mu Li, Alex Smola **發表：** 2026-06-17 ## 摘要 本文提出...
Towards Scalable Customization and Deployment of Multi-Agent Systems for Enterprise Applications
# Towards Scalable Customization and Deployment of Multi-Agent Systems for Enterprise Applications **作者：** Paresh Dashore, Shreyas Kulkarni, Uttam Gurram, Nadia Bathaee, Kartik Balasubramaniam, Genta...
Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting
# Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting **作者：** Guillermo Llopis **發表：** 2026-06-09 ## 摘要 企業的 AI 推理服務（API 訂閱、企業聊天工具、嵌入式 ...
Toward an AI-Powered Computational Testbed for Workforce Policy
# Toward an AI-Powered Computational Testbed for Workforce Policy **作者：** Sumer S. Vaid, Ashley V. Whillans **發表：** 2026-05-18 ## 摘要 AI 正大規模改變知識型工作，但企業缺乏工具來預測員工在組織變革中的心理與行為反應。本文提出「動態員工 Agent」概念——結合...
Understanding How Enterprises Adopt the Model Context Protocol for LLM-Driven Software Engineering
# Understanding How Enterprises Adopt the Model Context Protocol for LLM-Driven Software Engineering **作者：** Kehui Chen, Yicheng Sun, Jacky Keung, Zhenyu Mao, Xiaoxue Ma **發表：** 2026-06-08 ## 摘要 Mo...
A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents
# A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents **作者：** Krti Tallam **發表：** 2026-06-10 ## 摘要 當 AI Agent 在企業環境中執行自動化工作流程時，傳統的資料邊界安全模型不再適用。本文提出一個运行时治理的參考架構，將管控面拆解...
The Integrator Advantage: Controlled Agentic AI for Small and Medium-Sized Companies
# The Integrator Advantage: Controlled Agentic AI for Small and Medium-Sized Companies **作者：** Christopher Koch, Joshua A. Wellbrock **發表：** 2026-06-15 ## 摘要 本文探討 Agentic AI 對中小企業的價值，主張近期價值不在於完全自動化或...
Less Context, Better Agents: Efficient Context Engineering for Long-Horizon Tool-Using LLM Agents
# Less Context, Better Agents: Efficient Context Engineering for Long-Horizon Tool-Using LLM Agents **作者：** Abhilasha Lodha, Mahsa Pahlavikhah Varnosfaderani, Abir Chakraborty, Abhinav Mithal **發表：**...
To Share or Not to Share: Orchestrating Trustworthy Data in Global Value Chains
# To Share or Not to Share: Orchestrating Trustworthy Data in Global Value Chains **作者：** Han-Teng Liao, Chang-Yi Kao **發表：** 2026-06-11 ## 摘要 歐盟碳邊境調整機制（CBAM）即將上路，全球半導體價值鏈面臨「監管透明度」與「數據主權」的結構性矛盾。本文提...
Agentic AI 架構模式
# Agentic AI 架構模式 Agentic AI（自主 AI 代理）正在從單一 LLM 呼叫進化為具備規劃、工具使用與多步推理能力的自主系統。本文整理四種主流架構模式，並說明各自的適用場景與設計取捨。 ## 1. ReAct（Reasoning + Acting） ReAct 由 Yao et al. (2022) 提出，核心思路是將推理（Reasoning）與行動（Acting）交...
LLM 模型比較指南：主流模型強項與適用場景
# LLM 模型比較指南：主流模型強項與適用場景 截至 2026 年中，大語言模型市場已趨成熟，各模型在能力、成本、速度上呈現明顯分化。本文針對五個最具代表性的模型進行系統性比較，協助企業在 AI 選型時做出更明智的決策。 ## GPT-4o（OpenAI） OpenAI 的旗艦多模態模型，2025-2026 年歷經多次迭代，已成為事實上的業界標竿。 | 項目 | 內容 | |------...
RAG 技術深入：從 Naive RAG 到 Advanced RAG
# RAG 技術深入：從 Naive RAG 到 Advanced RAG RAG（Retrieval-Augmented Generation）已成為 LLM 落地的關鍵架構，解決了模型知識截止、幻覺與領域專業不足三大痛點。本文從 Naive RAG 出發，一路探討到 Advanced RAG 的核心技術。 ## Naive RAG 的架構與局限 Naive RAG 遵循經典的三段式流程：...
Shadow AI 資安防護：風險、偵測與治理框架
# Shadow AI 資安防護：風險、偵測與治理框架 Shadow AI（影子 AI）指的是未經 IT 或資安部門授權、未受企業治理規範約束的 AI 使用行為。隨著 LLM 服務日益普及——員工可直接在 ChatGPT、Claude、DeepSeek 等平台上貼入公司資料——Shadow AI 已成為企業面臨的最嚴峻資安挑戰之一。 ## Shadow AI 的五大風險 ### 1. 資料外...
Orchestrating the Twin Transition in Multinational Corporations: Technology Roadmapping for Green and Digital Global Business Services
# Orchestrating the Twin Transition in Multinational Corporations: Technology Roadmapping for Green and Digital Global Business Services **作者：** Han-Teng Liao, Karen Ang **發表：** 2026-06-11 ## 摘要 本文將...
Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting
# Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting **作者：** Guillermo Llopis **發表：** 2026-06-09 ## 摘要 本文提出四層級框架，將 AI 推論服務（API 訂閱、企業聊天...
Why Memory Components Fail: Eight Years of License and Sustainability Events in Open-Source Data Infrastructure
# Why Memory Components Fail: Eight Years of License and Sustainability Events in Open-Source Data Infrastructure **作者：** Dmitrii Dmitrenko **發表：** 2026-06-05 ## 摘要 本文指出 2026 年初五大 Agent Memory 調查論文均...
AI Agents for Sustainable SMEs: A Green ESG Assessment Framework
# AI Agents for Sustainable SMEs: A Green ESG Assessment Framework **作者：** Viet Trinh, Tan Nguyen, Minh-Huyen Phan, Quan Luu **發表：** 2026-04-05 ## 摘要 本文提出一個 AI 驅動的框架，用於評估歐洲中小企業的 ESG 表現。第一階段以 Flash E...
Prime Agent 架構分析 — 對 Hermes 的參考價值
> 分析日期：2026-08-09｜來源：github.com/PrimeIntellect-ai/prime-agent（v0.7.1，10k stars，MIT） > 方法：下載 release tgz，閱讀 dist/core/*.js 原始碼 + docs/architecture.md + skills/ 範例 ## 架構總覽 ``` TUI/headless clients ──►...
Prime Agent：Self-Improving RLM Agent（Prime Intellect 開源專案）
> 來源：https://github.com/PrimeIntellect-ai/prime-agent > 存入：2026-08-09｜分類：技術深讀 / AI Agent 架構 > 狀態：10k stars、988 forks、MIT License、v0.7.1（2026-08-07）、4,480 commits ## 一句話 開源的 coding/research agent，以「遞...
技術深讀
# 🔧 技術深讀 > AI 技術的深度研究筆記，從架構原理到實作細節，讓 CJ 哥在客戶面前說得出技術底蘊。 ## 📌 本領域重點 - Agentic AI 架構與設計模式 - RAG（檢索增強生成）技術深入 - LLM 模型比較與選型指南 - AI 資安與 Shadow AI 防護 - MCP / ACP 協議研究 - 影片生成與多模態技術 ## 📂 子主題 - Agentic ...