Daimler Truck CIO Scraps Point Solutions as AI Fails to Deliver Gains

因AI未能带来实际效益 戴姆勒卡车CIO弃用单点方案 Lutz Beck表示，81%的企业AI部署对底线利润毫无影响，因其构建的是孤立项目而非统一的企业基础。 Lutz Beck详细阐述了工业领域从孤立的传统自动化向自主且由AI驱动的端到端业务流程的转型。他指出，执行速度正在取代静态的行业领域知识成为核心竞争驱动力，同时披露81%的企业未从其AI举措中获得任何底线收益。为避免陷入“PoC（概念验证）坟场”，Beck提出了五项核心架构原则，涵盖更严格的治理、清洁的数据基础、流程重塑以及业务部门的所有权。最终，他展示了一种目标架构，主打通过单一统一的编排层来管理遍布全企业的各类专用数字化智能体。 各机构正从传统自动化（脚本、RPA）向AI增强型智能体系统转型，实现人类与数字员工并肩协作。 (0:08) 与早期旨在削减工作岗位的RPA应用不同，智能体AI是将具体任务委派给数字员工，而非取代整个人类岗位。 (3:22) 采用智能体模式使运营重心从孤立任务转向业务成果，从静态代码转向自适应系统，并从人工审批转向异常处理。 (4:02) 行业领域知识正迅速同质化与平庸化；制造业及工业领域的竞争优势如今由执行速度决定。 (4:42) IT部门及CIO的角色正在从被动支持职能，转变为企业的核心平台基座与转型办公室。 (6:20)

According to Lutz Beck, CIO of Daimler Truck North America, enterprise AI initiatives fail to generate measurable financial value when deployed as isolated experiments; organizations must instead build an integrated foundation with unified orchestration and end-to-end process redesign. The Enterprise AI Adoption and Return Gap (7:47) Despite widespread experimentation across corporate environments, Beck highlights a severe disconnect between technical deployment and tangible financial returns. Industry research cited in the presentation indicates that although 88% of companies have AI in production in at least one enterprise function, 81% see zero bottom-line gain or measurable P&L impact. As a result, only 6% of enterprises currently qualify as true value creators. Beck observes that many organizations manage AI like an academic research project rather than a commercial engine. Pursuing disconnected point solutions without an enterprise-wide foundation leads organizations directly into a 'proof-of-concept graveyard' characterized by cultural resistance, low internal adoption, lack of business ownership, and an inability to demonstrate return on investment. Moving from Task Automation to Operational Autonomy (4:02) The operational shift from basic automation to an agentic enterprise fundamentally changes how work is executed. While early Robotic Process Automation (RPA) largely operated as a head-cutting exercise attempting to remove human jobs via rigid scripts, agentic systems delegate discrete tasks to adaptive digital workers while humans manage exceptions. Beck notes that currently only about 15% of day-to-day industrial decisions are made autonomously, but establishing even this level of autonomy across legacy-heavy manufacturing provides a substantial competitive advantage. Furthermore, Beck asserts that AI models are rapidly commoditizing traditional domain knowledge in sectors like trucking and manufacturing. Because technical expertise is no longer a moat, organizational execution speed has become the primary competitive differentiator. Research cited indicates that 33% of enterprise applications will soon embed agentic AI capabilities. Foundational Principles for Scalable Enterprise Architecture (10:03) To escape the proof-of-concept trap, Daimler Truck North America relies on five core operating pillars. The operating model assigns technology framework ownership to IT through a plug-and-play architecture, while business units directly own and drive the AI initiatives. Data architecture must also be overhauled; Beck warns that simply transferring raw or manipulated mainframe data into modern repositories like Snowflake or Databricks cannot support dependable AI systems. Additionally, high-performing organizations treat AI capabilities as internal products to retain proprietary intellectual property rather than outsourcing all technical knowledge to vendors. Value chains must be redesigned from the ground up rather than overlaying AI on flawed legacy processes, and projects must justify themselves through concrete top- or bottom-line financial gains while monitoring ongoing compute and runtime overhead. Unified Orchestration and Preserving Human Judgment (16:27) Daimler Truck North America’s target architectural vision avoids deploying isolated, department-specific orchestrators. Instead, Beck outlines a model where a single, central orchestration layer sits directly above an enterprise foundation model and data backbone to coordinate specialized agents across all business functions. Within this architecture, Beck maintains that clear boundaries must separate digital and human roles. While digital agents excel at repeatable task execution, human workers remain irreplaceable for critical thinking, ethical judgment, and relational trust. Preparing the workforce to collaborate effectively alongside digital workers therefore requires deliberate, enterprise-wide upskilling. Beck concludes that long-term enterprise winners will not be the organizations with the most individual AI features, but those that establish an integrated, autonomous system. He warns that legacy incumbents who fail to redesign core workflows risk disruption from faster competitors who can decouple and capture high-value operations.

戴姆勒卡车北美公司（Daimler Truck North America）首席信息官Lutz Beck指出，当前多数企业的人工智能实践因缺乏整体架构而陷入低效的“概念验证墓地”；企业必须从局部的任务自动化转向由单一体编排层驱动的智能体系统，通过自底向上的流程再造和严格的损益考核来实现真正的业务自主化。 概念验证墓地与企业AI落地困局 (7:47) Beck对当前企业级AI的应用现状给出了严峻的行业评估。他指出，许多企业中的AI项目更接近学术研究，而非实际业务增长的引擎。根据其引用的统计数据，虽然已有88%的企业在至少一个业务职能中将AI投入生产运营，但高达81%的企业并未借此获得净利润提升（bottom-line gain），对损益表（P&L）毫无实质性贡献，仅有6%的企业能够被视作真正的价值创造者。 造成这种落差的主因是企业普遍受困于孤立的用例驱动模式，缺乏易于在全企业范围内推广的基础架构模型。没有系统性的顶层设计，单点式的解决方案极易引发组织文化抵触、实际采纳率低下、投资回报率模糊以及业务部门缺乏归属感等问题，最终导致大量项目沦入停滞不前的“概念验证（PoC）墓地”。 从脚本自动化到智能体协同的范式演进 (4:02) Beck回顾了技术从基础脚本、机器人流程自动化（RPA）、预测性机器学习、生成式AI到智能体系统的演进历程。他认为，早期的RPA本质上是以削减员工岗位为导向的局部操作，而智能体系统则是将具体工作任务委托给数字员工，实现人类与数字劳动力的协同作业。 这种演进推动了三项核心运营范式的转变：从关注单项任务执行转向交付整体业务成果，从硬编码的静态规则转向自适应学习系统，以及从人工微观逐级审批转向仅由人类处理异常的自主运转。Beck指出，目前工业领域日常决策中仅约15%实现了自主化，但哪怕达到这一水平也能在重资产传统领域形成巨大壁垒；同时，AI正在使静态领域知识大众化，未来的竞争优势将取决于执行速度，行业研究亦预测将有33%的企业级应用很快内嵌智能体能力。 架构智能体企业的五大核心支柱 (10:03) 为规避概念验证失败的风险，戴姆勒卡车北美公司提出了转型必须遵循的五项原则：第一是运营模型与治理机制，IT部门必须由传统的后勤保障转变为核心平台与转型办公室，负责搭建即插即用的技术架构与安全护栏，并由业务部门全权负责AI举措的业务落地；第二是夯实企业数据与知识底座，强调单纯将原始数据迁移至Snowflake或Databricks并不代表架构合格，尤其需要整治遗留大型机中经过人为修改、无法被AI有效利用的数据。 第三是将AI能力视作企业内部产品进行长期建设，以沉淀核心技术知识产权并避免过度依赖外部供应商；第四是彻底重构端到端业务流程，反对直接在存在缺陷的旧流程上硬套AI技术；第五是紧扣商业成果与权责归属，排斥所谓“规避成本（avoided costs）”等软指标，坚持以损益表中的增量收益为准绳，并严格监控以“日常AI（Everyday AI）”算力开销为代表的系统运行资源消耗。 统一编排架构与人机协同边界 (16:27) 在目标架构规划上，Beck描绘了自主型企业的技术形态：在跨职能业务价值层与底层企业基础模型/数据底座之间，必须架设一个“单一体编排层（Single Orchestration Layer）”。该层统一管理并调度跨越不同业务领域与职能部门的所有专用数字智能体，彻底摒弃各部门分别采购或开发分散编排器的孤岛模式。 在明确自动化目标的同时，Beck划定了AI的能力边界。数字智能体擅长任务执行，但在批判性思维、伦理道德判断以及人际信任建立方面无法替代人类，因此推动现有员工的技能升级转型是落地智能体架构的必要前提条件。 Beck总结指出，未来的行业主导者不会是那些仅仅罗列AI功能清单的公司，而是成功构建集成化自主系统的企业；鉴于领域知识正在被技术快速平民化，若传统工业巨头不能主动承担构建智能体架构的艰巨工作并重塑核心流程，其高价值业务势必面临被敏捷新兴竞争对手侵蚀和颠覆的风险。

## Background 该内容由 Ai4 于 YouTube 平台发布，记录了戴姆勒卡车北美公司（Daimler Truck North America）CIO Lutz Beck 的主题演讲。视频主要探讨企业如何从孤岛式自动化跨越至具备自主规划、决策与执行能力的智能体系统（Agentic Enterprise），并深度阐述了 IT 领导者与 CIO 在这场企业级架构重塑与业务赋能转型中的核心使命。 ## TL;DR 戴姆勒卡车北美公司（DTNA）指出，AI 正在从以削减人头为导向的传统 RPA 自动化全面跨越至任务委托与人机协同的智能体（Agentic）时代。面对 81% 的企业 AI 投资无法改善财务底线收益并深陷“PoC 墓地”的现实，企业必须立足于即插即用架构、数据底座、AI 内部产品化、流程再造以及硬性损益表回报五大基石。随着底层模型与行业领域知识日益商品化，执行与迭代速度已成为决胜关键，IT 部门也正在从传统支持角色演变为平台变革中枢。DTNA 规划的终极蓝图依托全局唯一的中央编排层来调度跨职能自主智能体，在人类把控伦理与信任的同时实现端到端业务闭环。 ## Atmosphere & style 演讲整体呈现出大型工业制造企业特有的严谨、务实与直面现实的犀利风格，摒弃了空洞的技术炒作，直击行业“PoC 墓地”与虚假 ROI 的痛点。主讲人语调冷静沉着且极具战略前瞻性，以清晰的逻辑框架和硬核的架构推演展现了传统制造巨头主动求变的决心。整场分享节奏紧凑、危机感与使命感并存，既有对遗留系统和算力成本的现实审视，又传递出以“速度战胜领域知识”颠覆传统运营范式的坚定信念。 ## Key takeaways - [00:08](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=8s) 工业企业正从传统的孤岛式自动化（Automation）迈向具备自主规划、执行与纠偏能力的智能体自主系统（Autonomy）。 - [00:08](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=8s) 与 RPA 传统的削减岗位导向不同，智能体模式是将具体任务委托给数字员工，通过重构工作流实现人机协同作业。 - [04:02](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=242s) 企业核心竞争力正在发生范式转移，基础模型与领域知识逐渐商品化，“执行与迭代的速度”战胜了单纯的领域知识积累。 - [04:02](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=242s) IT 部门的角色被彻底重定义，从传统的底层支持与运维部门转变为统筹全局技术底座的“企业变革办公室”。 - [07:47](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=467s) 工业界 AI 普遍陷入“PoC 墓地”陷阱，81% 的企业未能获得实质财务回报，仅 6% 的先锋企业成功跑通价值闭环。 - [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) 智能体企业的治理原则必须明确“业务拥有 AI 业务所有权，IT 负责底座赋能”，并采用即插即用的动态解耦架构。 - [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) AI 必须作为企业内部核心产品进行全生命周期管理，保留自主核心能力，并彻底夯实底层数据与知识底座。 - [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) 企业必须推倒重塑旧流程，并以损益表的硬性财务指标衡量 AI 价值，同时严格管控算力与 Token 运行成本。 - [16:27](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=987s) 人类在伦理道德判断与信任构建上不可替代，在部署智能体系统的同时必须对员工进行系统性的技能升级与重塑。 - [16:27](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=987s) 智能体企业的终极架构依托于唯一的“中央编排层”，全局统筹调度各职能智能体以实现端到端业务闭环。 - [16:27](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=987s) 未来的核心竞争壁垒不在于堆砌了多少孤立的 AI 功能，而在于能否将 AI、高质量数据与业务流程有机融为一体。 ## Detailed walkthrough ### [00:08](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=8s) 工业界现状：从孤岛式自动化迈向自主系统 在传统工业制造中，生产线与车间层面早已实现了高度的机械与流程自动化，然而在行政、管理及业务流程后台，企业仍然严重依赖单点的人工投入。这些行政环节缺乏端到端的拉通，各类软件和工具彼此割裂，形成了严重的业务孤岛。戴姆勒卡车北美公司（Daimler Truck North America, DTNA）明确指出，企业所追求的“新世界”必须建立在 AI 赋能的流程和智能系统之上。在这一体系中，人类与机器协同作业，企业内部甚至会拥有数量超过人类员工的“数字员工（Digital Workers）”，共同驱动运营效率和实际业务结果。 在工业企业落地技术的核心原则是“每一项投入都必须创造实际价值”。企业没有时间也没有资本进行无意义的试水或玩票性质的技术探索，一切都必须以成本与业务价值为导向。过去几年企业并非没有在自动化上投入，但当下的关键转型在于——彻底从单纯的“自动化（Automation）”跨越到“自主化（Autonomy）”。自主系统的本质在于具备自主规划（Plan）、执行（Act）并在偏离既定目标时自主纠偏（Correct）的能力。 ```mermaid flowchart LR accTitle: 从传统自动化向智能体自主化的演进 subgraph S1["传统模式（自动化）"] direction TB A1["脚本与 RPA 工具"] --> A2["单点/孤岛式流程"] A2 --> A3["以裁员为导向的削减"] end subgraph S2["演进过渡"] direction TB B1["预测型机器学习"] --> B2["生成式 AI 浪潮"] end subgraph S3["未来模式（自主化）"] direction TB C1["智能体系统 (Agentic)"] --> C2["自主规划/执行/纠偏"] C2 --> C3["数字员工与人类协同"] end S1 --> S2 --> S3 ``` - 技术路径的演进经历了四个阶段：从最基础的脚本与 RPA（机器人流程自动化），到预测型机器学习（Predictive ML），再到生成式 AI（GenAI），如今全面迈向智能体系统（Agentic AI）。 - 过去 RPA 模式的本质往往是“裁减岗位/人头（Head cutting exercise）”，在业务端并没有取得真正的长效成功；而 Agentic 模式是“将具体任务委托给数字员工”，并非直接淘汰整个岗位，而是重构工作方式。 --- ### [04:02](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=242s) 范式转变：速度战胜领域知识与 IT 领导者角色重构 智能体时代的到来为企业带来了根本性的范式转移：工作重心从“关注具体任务”转向“追求最终结果”；系统架构从“编写静态代码”转向“自适应演化系统”；管理逻辑从“逐层人工审批”转向“智能体自主决策与人工异常处理”。尽管当前工业界由系统自主做出日常决策的比例可能仅在 15% 左右，但这恰恰是构筑未来竞争壁垒的关键空间。行业内正在发生一个巨大的认知颠覆：基础模型正在高度商品化（Commoditizing），过往引以为傲的“行业领域知识（Domain Knowledge）”在很大程度上也逐渐变成了通用的大宗商品。 在传统制造企业中，员工往往习惯以掌握深厚的领域知识来定义自身价值，然而未来的核心竞争力在于“执行与迭代的速度”。随着预测约 33% 的企业级应用将深度嵌入智能体 AI，技术领导者与 IT 部门的角色必须被重新定义。在过去，IT 仅仅被视作一个提供应用运维、基础设施保障、项目交付以及安全防护的传统支持部门（Support Function）；而在智能体企业架构中，IT 演变为了整个企业的平台中枢与“变革办公室（Transformation Office）”。 - **范式转变的核心维度**： - 从关注任务（Tasks）转向交付成果（Outcomes）。 - 从静态代码（Static Code）转向自适应系统（Adaptive Systems）。 - 从逐级审批（Approval）转向异常治理（Exception Handling）。 - 速度优先原则（Speed over Domain Knowledge）：当底层模型与领域认知逐渐泛化，谁能以最快速度将技术转化为落地能力，谁就能抢占先机。 - IT 部门正在从传统的底层支撑角色，转变为企业级架构设计师与平台赋能者，统筹全局技术底座以赋能业务端自主构建解决方案。 --- ### [07:47](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=467s) 工业界残酷的现实检验与 PoC 墓地 审视当下工业界落地 AI 的真实数据，表面的繁荣与底层的业务回报之间存在着触目惊心的鸿沟。企业如果仅仅停留在单个用例（Use Cases）的修修补补上，AI 投资在本质上就等同于“昂贵的科研实验”，而无法为企业损益表带来实质贡献。 | 统计指标 | 占比数据 | 行业现状解读 | |:--- |:--- |:--- | | **已在生产环境中部署 AI 的企业** | **88%** | 绝大多数企业至少在一个业务职能中上线了 AI 应用。 | | **底线收益与损益表无实质改善的企业** | **81%** | 绝大部分企业未能通过 AI 创造出可量化的财务底线收益。 | | **真正的价值创造者（True Value Creators）** | **6%** | 仅极少数先锋企业成功跑通价值闭环，将 AI 转化为核心生产力。 | 这一脱节的核心根源在于：市场上至今缺乏开箱即用、能轻松在企业全局层面扩展的“企业级基础模型（Enterprise Foundation Model）”。这导致企业陷入了重重泥潭： - **概念验证墓地（PoC Graveyard）**：大量项目停留在试点验证阶段，无法跨越到生产规模化。 - **局部效率陷阱**：仅能在极其孤立的细分环节带来微弱效率提升，无法拉通全局价值。 - **文化与心智阻力**：AI 落地从来不是单纯的技术问题，而是涉及组织文化与思维模式转型的系统性工程。 - **投资回报模糊与权责缺失**：业务用例缺乏明确的 ROI 衡量标准，碎片化解决方案泛滥且无人对其最终成效负责。 --- ### [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) 架构智能体企业的五大基石 为了打破“PoC 墓地”的怪圈，DTNA 总结出了构建智能体企业必须具备的五大核心支撑体系。这套体系明确了业务与 IT 之间的责任边界：“业务部门拥有 AI 业务所有权，IT 部门负责赋能并搭建基础设施（Business owns AI, IT enables AI）”。 ```mermaid flowchart TD accTitle: 智能体企业的五大支撑基石 subgraph P1["运营与架构治理"] A["业务与 IT 深度协同"] B["即插即用架构 (Plug-and-Play)"] C["Best-of-Breed 动态选型"] end subgraph P2["数据与产品基建"] D["企业级数据与知识底座"] E["AI 作为内部产品体系运作"] end subgraph P3["流程重塑与价值度量"] F["端到端业务与流程彻底重塑"] G["损益表级硬性价值度量与算力成本控制"] end P1 --> P2 --> P3 ``` 1. **运营模型与架构治理（Operating Model and Governance）**：建立业务与 IT 的紧密协同模型，树立明确的架构原则与护栏机制。DTNA 坚持采用“即插即用（Plug-and-Play）架构”，所有接插件与接入框架由内部严格把控，允许随时将业内“最优（Best of Breed）”方案接入；一旦某家供应商不再处于顶尖水平，便可立即替换出局。 2. **企业级数据与知识底座（Enterprise Data & Knowledge Foundation）**：没有扎实的数据底座，AI 规模化便无从谈起。很多企业误以为把数据丢进 Snowflake 或 Databricks 就万事大吉，但实际上大量传统制造企业依然运行着陈旧的大型机（Mainframe），其数据存在严重的任意篡改和质量缺陷，无法直接用于支撑高质量 AI。 3. **AI 作为内部产品（AI as an Internal Product）**：AI 必须作为内部产品来进行生命周期管理以沉淀复合价值。核心技术框架与自主控制权必须牢牢掌握在自己手中，绝不能直接整体外包给外部供应商而丧失关键技术积累。 4. **工作流与业务流程重塑（Workflow or Process Redesign）**：高效能组织绝不仅仅是在现有用例上缝补，而是敢于推倒并重塑旧有流程。审视端到端价值链，承认“过去带来成功的旧流程已无法适应明天”，在底层遗留系统错综复杂的情况下坚决推动流程再造。 5. **明确的业务成果与责任制（Business Outcomes and Ownership）**：拒绝所谓的“避免成本（Avoided costs）”等纸面虚假指标，AI 必须带来真实的营收增长（Top line）或利润改善（Bottom line）。同时在全公司推行日常 AI 工具（如 DTNA 的 "Everyday AI"）时，必须严格监控算力消耗与运行成本（Cost to run），避免 Token 和算力成本失控飙升。 --- ### [16:27](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=987s) AI 的能力边界与未来目标架构：单一编排层驱动的智能体企业 在积极推进技术落地的同时，企业必须对 AI 的能力边界保持清醒。RPA 是为了削减岗位，而 AI 则是为了分担特定任务。人类员工在“伦理道德判断（Ethical Judgment）”和“建立信任关系（Relational Trust）”等维度具有不可替代的核心价值。企业在推进 Agentic 系统的同时，必须系统性地对员工进行技能重塑与提升（Upskill），帮助他们承担更高级别的职责。 在 DTNA 规划的最终目标愿景中，智能体企业将依托一个全局统一的“中央编排层（Central Orchestration Layer）”运转： ```mermaid flowchart TD accTitle: 智能体企业目标架构 subgraph Top["跨职能业务执行层"] direction LR U1["生产制造"] --- U2["供应链管理"] --- U3["行政与财务"] end subgraph Mid["核心大脑：中央统一编排层"] ORCH["单一企业级编排中心 (Orchestration Layer)<br/>· 调度跨职能自主智能体<br/>· 端到端任务闭环执行"] end subgraph Bot["底座支撑"] EFM["企业级基础模型与知识库 (Enterprise Foundation Model)"] end Top <--> ORCH ORCH <--> Bot ``` - **单一编排中枢**：未来企业内部不应存在两套、三套割裂的调度引擎，而是依靠唯一的编排层来统筹所有智能体的运转，无论该智能体属于何种业务板块或执行何种职能。 - **跨职能自主闭环**：编排层直接调度各个智能体，打通端到端的跨职能业务用例，完全自主地规划并执行完整业务任务。 - **最终竞争优势**：未来的赢家绝不是那些堆砌了最多零散 AI 功能（AI features）的公司，而是成功将 AI、数据与业务流程有机整合为一体的现代化企业。 - **时代紧迫性**：由于行业领域知识正在迅速被模型化与商品化，如果传统企业不主动求变、加速构建自身的智能体底座，外部新兴力量将会快速渗透并蚕食企业的各项核心业务，最终将传统企业挤压边缘化。数字化与自主化是一场艰苦的长期转型，但唯有敢于重构的企业才能赢得未来。 ## Notable quotes > [00:08](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=8s) 每一项投入都必须创造实际价值 — Lutz Beck > [00:08](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=8s) 过去 RPA 模式的本质往往是“裁减岗位/人头（Head cutting exercise）”，在业务端并没有取得真正的长效成功；而 Agentic 模式是“将具体任务委托给数字员工”，并非直接淘汰整个岗位，而是重构工作方式。 — Lutz Beck > [04:02](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=242s) 基础模型正在高度商品化（Commoditizing），过往引以为傲的“行业领域知识（Domain Knowledge）”在很大程度上也逐渐变成了通用的大宗商品。 — Lutz Beck > [04:02](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=242s) 在智能体企业架构中，IT 演变为了整个企业的平台中枢与“变革办公室（Transformation Office）”。 — Lutz Beck > [07:47](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=467s) 企业如果仅仅停留在单个用例（Use Cases）的修修补补上，AI 投资在本质上就等同于“昂贵的科研实验”，而无法为企业损益表带来实质贡献。 — Lutz Beck > [07:47](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=467s) AI 落地从来不是单纯的技术问题，而是涉及组织文化与思维模式转型的系统性工程。 — Lutz Beck > [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) 业务部门拥有 AI 业务所有权，IT 部门负责赋能并搭建基础设施（Business owns AI, IT enables AI）。 — Lutz Beck > [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) AI 必须作为内部产品来进行生命周期管理以沉淀复合价值。核心技术框架与自主控制权必须牢牢掌握在自己手中，绝不能直接整体外包给外部供应商而丧失关键技术积累。 — Lutz Beck > [10:03](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=603s) 拒绝所谓的“避免成本（Avoided costs）”等纸面虚假指标，AI 必须带来真实的营收增长（Top line）或利润改善（Bottom line）。 — Lutz Beck > [16:27](https://www.youtube.com/watch?v=dP5yDdMKNuQ&t=987s) 未来的赢家绝不是那些堆砌了最多零散 AI 功能（AI features）的公司，而是成功将 AI、数据与业务流程有机整合为一体的现代化企业。 — Lutz Beck ## Mentioned Lutz Beck, 戴姆勒卡车北美公司 (Daimler Truck North America / DTNA), Ai4, Snowflake, Databricks, RPA, 生成式 AI (GenAI), 智能体 AI (Agentic AI), Everyday AI ## Video description Lutz Beck - CIO, Daimler Truck North America AI is advancing rapidly—from automation to autonomy. Agentic AI systems can plan, decide, and execute across entire business processes, fundamentally changing how companies operate, how processes are designed, and how work gets done. In this session, we’ll explore the central role IT and CIOs play in this shift, how to architect the agentic enterprise, and how to enable the business throughout this transformation. [Source](https://www.youtube.com/watch?v=dP5yDdMKNuQ)

[0:08] Lutz Beck: 每一项投入都必须创造实际价值 [0:08] Lutz Beck: 过去 RPA 模式的本质往往是“裁减岗位/人头（Head cutting exercise）”，在业务端并没有取得真正的长效成功；而 Agentic 模式是“将具体任务委托给数字员工”，并非直接淘汰整个岗位，而是重构工作方式。 [4:02] Lutz Beck: 基础模型正在高度商品化（Commoditizing），过往引以为傲的“行业领域知识（Domain Knowledge）”在很大程度上也逐渐变成了通用的大宗商品。 [4:02] Lutz Beck: 在智能体企业架构中，IT 演变为了整个企业的平台中枢与“变革办公室（Transformation Office）”。 [7:47] Lutz Beck: 企业如果仅仅停留在单个用例（Use Cases）的修修补补上，AI 投资在本质上就等同于“昂贵的科研实验”，而无法为企业损益表带来实质贡献。 [7:47] Lutz Beck: AI 落地从来不是单纯的技术问题，而是涉及组织文化与思维模式转型的系统性工程。 [10:03] Lutz Beck: 业务部门拥有 AI 业务所有权，IT 部门负责赋能并搭建基础设施（Business owns AI, IT enables AI）。 [10:03] Lutz Beck: AI 必须作为内部产品来进行生命周期管理以沉淀复合价值。核心技术框架与自主控制权必须牢牢掌握在自己手中，绝不能直接整体外包给外部供应商而丧失关键技术积累。 [10:03] Lutz Beck: 拒绝所谓的“避免成本（Avoided costs）”等纸面虚假指标，AI 必须带来真实的营收增长（Top line）或利润改善（Bottom line）。 [16:27] Lutz Beck: 未来的赢家绝不是那些堆砌了最多零散 AI 功能（AI features）的公司，而是成功将 AI、数据与业务流程有机整合为一体的现代化企业。

[0:08] Lutz Beck: organizations may soon employ more digital workers than human employees in certain domains. [0:08] Lutz Beck: companies simply do not have the time or resources to "play around." [4:02] Lutz Beck: competitive advantage is now dictated by execution speed over static domain knowledge. [4:02] Lutz Beck: 33% of enterprise applications will soon embed agentic AI capabilities. [7:47] Lutz Beck: 88% of companies have AI in production in at least one enterprise function. [7:47] Lutz Beck: 81% of companies generate zero bottom-line gain and no measurable P&L impact. [7:47] Lutz Beck: Only 6% of companies qualify as true value creators. [7:47] Lutz Beck: Without a cohesive architecture, organizations stall out in a "PoC (Proof of Concept) graveyard" [10:03] Lutz Beck: IT owns the underlying technology framework into which best-of-breed solutions are plugged, while the business owns and drives the AI initiatives. [10:03] Lutz Beck: High performers redesign core business processes from scratch rather than slapping AI on top of flawed legacy workflows. [16:27] Lutz Beck: While AI excels at task execution, humans are required for critical thinking, ethical judgment, and relational trust. [20:00] Lutz Beck: the ultimate winners in the market will not be the companies boasting the longest list of individual AI features, but those that successfully build an integrated, autonomous enterprise system.

[0:08] So let me let me try to bring some flavor out  of the industry uh onto the conference and what [0:16] what I mean with automation to autonomy right  um so when you when you look at it right [0:24] specifically in our industry right this that's  the old world Yeah, everybody is aware of that. [0:31] Everybody knows that this is how it's done,  right? And this is still some kind of uh topics [0:41] which we have even even though we have automation  achieved of course significantly in the production [0:48] and in other areas you still see certain things  like processes. It's more individual effort. It's [0:55] not end to end. It's not things disconnected tools  also and everything what we have and of course [1:03] we're working towards an automation uh but when  you look at production highly automated highly [1:12] uh done with tools and everything when you  look at all the administrative processes [1:18] it's siloed and it's not really working.  Now what we of course want is the new world [1:25] uh which is given with the AI  shift to the AI enhanced processes [1:32] intelligent systems right uh humans and machines  collaborate right we have digital workers we [1:39] probably have more digital workers than humans  in a certain sense right and we drive operational [1:46] efficiency and business results very important  everything what we're doing needs to have a value. [1:53] We're not playing around and we we don't have we  don't have the time to play around. So everything [1:58] what we have is costdriven thing and we need to  deliver the value. Now what is the shift here when [2:05] you look at it because it's not that we were not  doing anything over the past years in automation [2:12] but uh and this is what I'm coming to. We shift  basically everything from automation to autonomy. [2:21] Yeah. So we want to have systems that  plan act and of course correct towards [2:26] defined goals autonomously. So we have if  you look at it in the different dimension [2:33] of course you were having scripts RPA you were  having predictive ML then J gen GI was coming [2:42] everything was changing the world was changing.  It's not that we worked with AI not before but [2:48] there is and of course now everything is  agentic and everything needs to be agentic [2:55] and when you look at it what we see from different  uh dimensions that the automation was having its [3:05] topics and was helping us in certain sense but  the Gentic system will help us significantly [3:14] to change everything what we're doing not just in  production across the whole enterprise and across [3:20] the whole processes and this is something which we  looking at because we were not really successful [3:27] if we need to look into that with RPA because RPA  was a head cutting exercise. Yeah, agentic is not [3:39] head cutting. Agentic is we will shift tasks to  digital workers but we will not shift complete [3:48] jobs to digital workers and that's the big topic  and that's something which we are working on and [3:54] where we are looking and how we can do it and how  we should work on that one over the future. Now [4:02] why is that a big change for us?  So we go from tasks to outcomes. [4:09] We go from static code to adaptive systems.  We go from approval to exception handling. [4:17] Yeah. So hopefully we will be able to reach this  15% and you might say this is a very small number [4:25] which is there 15% there is much more but if  you look at the industry and if you look across [4:31] the industry day-to-day decisions are not made  autonomously by now and we need to continuously [4:37] working that and it will help us also to  change and get competitive advantages. But [4:44] one thing is clear and we heard that just  before here models are commod commoditizing. So [4:51] it's not more about your domain knowledge  because this is in a large sense commodity. [4:58] It's more about the speed and the speed what  we need to do and how we do. And then of course [5:05] we need to move speed over the domain  knowledge is a very important thing for us [5:09] because this will help us to be changing  significantly and we are in an industry which has [5:18] domain knowledge significantly and people  define themselves by the domain knowledge [5:23] and not by the speed. So this will change  now significantly how we do the things [5:28] and how we do it. You see here also uh 33% of  enterprise apps will be embedded at Gentic AI. Um [5:37] let's see uh and let's look into  that if this is really coming [5:43] when we look at that and when we look  at the change it requires also a change [5:51] in a sense for technology leaders because so far  and everything what was done in the automation [5:59] it in general traditional sense was a support  function. Yeah. suppliers sometimes as well. [6:08] Yeah. So of course you need to run and secure the  systems, you have the applications, you have the [6:14] infrastructure, you have that to you have the  road maps, you have the project, you have the [6:19] built and all that kind of topics, but at the  end of the day you were stuck in the old world [6:26] and everybody was looking to you in the old  world. Now when you look at the new world, [6:32] it is the platform and it is the foundation  and no one else is actually capable to do [6:40] and to bring that all together. Right? So  the foundational element need to be there [6:45] and you need to be more on the architecture role  than ever before because you design something [6:50] which is is holistic for the whole enterprise  and where you need to look into that. So this [6:55] is shifting significantly. So then you need  to bring in platforms, technology frameworks, [7:02] yeah, which you can give out to the users  that they can build their topics as well. [7:08] So and then you have of course uh enabling  factor as well. So we call it in the meanwhile [7:15] it is the transformation office of the enterprise  because you need to drive business and technology [7:24] uh changes and this is very important  and we need to look into that um [7:31] how we do it uh we need to prepare ourselves as a  function significantly for that but we also need [7:37] to look into how we are doing that. Now when  you look into it, let's do the reality check [7:47] and maybe these numbers are slightly higher  or slightly lower but everybody here in [7:55] the conference will probably say hey this is  much far too low and whatever but if you look [8:01] 88% of companies have AI in production in at  least one enterprise function which is good. [8:13] 81% of companies generate zero bottom line gain  with no measurable PL impact and 6% of companies [8:21] are true value creators. 6%. Yeah. So when you  look from an industry perspective, AI is research [8:32] because it's not generating the value which  we actually need on the industry side. Why? [8:39] Because we are still stuck  on use cases. There is no [8:49] enterprise foundation model which is easy to  adopt from an enterprise level. It's not existing. [8:57] I know there's companies out there they start  thinking about it and they start building that. [9:02] But this is actually what would scale AI  significantly if we would have the chance to do [9:08] that and have that uh enterprise foundational  model. So this is something where we hope [9:14] uh that we can look into that and have that.  So you see also local efficiency gains only [9:23] P graveyard. Yes. I don't know. I mean I can  I can just look Oops. I can just look at our [9:31] PC graveyard uh low adoption. There is still  a lot of resistance because AI is not about [9:39] the technology is about the cultural shift,  the mindset shift in the company. So this is [9:43] something which we need to have unclear return on  sales uh return on invest there is no ownership [9:51] and fragmented solutions. This is all stuff which  we looking in when we when we look at it right and [9:58] now what does it take? So we were looking  into it because we don't want to have [10:03] the PC graveyard or whatever. So we were  looking into it and say okay what do we need [10:12] and I start maybe from the right side from  my perspective operating model and governance [10:20] it's something which needs to be clear which  needs to be there in place and where you need [10:24] to have so you need to have business and IT  aligned you need to have a clear working model [10:30] you need to have a prioritization you have  guardrails in the architecture principles [10:35] you have a technology framework. That means  also you cannot do everything. How we define it [10:42] our architecture is a plugandplay architecture.  So everything what is in order to pluck [10:50] is ours is the technology framework but the  technology framework should be able to basically [10:56] give us a possibility to block any solution  into this framework which we will want to use. [11:02] What we also do is we take the best of breed. If a  company is not more delivering best of breed, they [11:08] are out, right? Because it doesn't make sense.  The finance or the budget model is of course still [11:15] something which we need to agree on because it's  based on value and it's not a license, right? So [11:20] this is things which we need to have and business  owns AI, it enables AI. Yeah, it's not. So they [11:28] need to drive that and this is still a thing. Then  the enterprise data and the knowledge foundation [11:36] big topic everybody knows that but if you  not have a data foundation you have no scale [11:45] and the data foundation there is not much  companies out there in the industry which were [11:50] working on the data foundation because just moving  data into a snowflake or in a data bricks will not [11:57] help you right there is a lot of other things  in in there. If you look across the industries, [12:04] how much mainframe we still have all over.  Now mainframe data is not necessarily the [12:12] most qualitative data because this is the main  frame gives you the chance to do to manipulate [12:17] everything what you want to do. So the data  you cannot really use if you want to scale AI [12:23] and what when you want to use that. So this is  something which needs to be in place as well. [12:29] Then AI as a internal product. Yeah. So  products compound value is something which [12:37] we need to look into as well. And of course  we need to steer it. We need to own that [12:42] in a certain sense. I cannot give it some  kind of a vendor and say just do it for me [12:48] because I will lose a lot of the knowledge. So  there must be a thing where we say hey we know the [12:55] technology framework. We know how to do it. We can  do it by ourselves and of course it doesn't matter [13:02] whatever startup or company is helping us to scale  that then there is a workflow or process redesign. [13:12] If you really want to do big PS and if you really  want to change the enterprise you need to change [13:19] the processes. You need to look at your current  model. You need to look at your process model. [13:24] You need to look at your business model. You  need to look at your end to end value chain [13:29] and you need to say okay everything  what I did till today was successful [13:35] and was helping us tomorrow it will not more.  So we need to change that. We need to look in [13:41] and this is a big task because at the end of the  day you need to see how you can do it and how you [13:47] can flow that because at the end nobody knows  the processes especially when you are sit on [13:53] old legacy technology it's very difficult to find  that so but this is a prerequisite for us as well [14:00] and I think high performer redesign the processes  and not just do use casebased something and then [14:07] the last point which is veryant important as  well is the business outcomes and the ownership. [14:15] Yeah. So clear value  proposition and business case. [14:21] I have a lot of AI use cases which  we implemented over the time and [14:30] there is a lot of business cases. Now there is  business cases which showing a revenue increase [14:40] which showing we would have an  efficiency increase but it's more [14:45] some kind of uh I call it avoided costs.  So there is not real efficiencies. Yeah. [14:52] And this is something we can as an industry not do  go on cases which are not helping us significantly [14:59] increased efficiency or bottom line or top line  something needs to come. If this is not there [15:07] doesn't make sense right and this is one big  topic and then you need the business owners [15:13] and you need to measure uh on the  enterprise level not on a pilot level [15:19] with the measurement is coming also and this is  uh increasing of course here as well we need to [15:25] look into what is the cost to run these solutions.  Yeah, what is the consumption when we do all that? [15:34] We have a we have something which is called  everyday AI for all our users in the company [15:41] but I need to measure what they are  doing and I need to educate them how [15:44] they do things right because otherwise  my consumption goes through the roof [15:49] and this is something which we have to  work as well. So that's our principles [15:55] uh to look in at what we are doing but at  the bottom line is always it needs to be a [16:00] value driver. If it's not driving value if it's  not significantly helping us to change topics [16:07] it's nice but I need to look into it then like  I'm doing with some computing stuff as well right [16:19] because it's not helping us. So we need  to see how we can do it. Then of course [16:27] we also need to be clear what AI cannot  do because it is help for us also helping. [16:37] So there is still and that's the reason  why I said before RPA was a position cut. [16:45] AI is a task cut but nothing else because you  need to humans still for doing certain things [16:52] which you don't have in place. Yeah. So there is  this three elements. I take the middle one because [16:58] this is coming more and more into the picture  for for companies is the ethical judgment. Um [17:06] you can also take the relational trust that's  stuff which is coming alongside with everything [17:12] what we are doing and how we are doing things but  this is very important for us to always consider [17:17] that and not just go on the topics and say hey we  can we can do that or we cannot do that. So this [17:24] is elements we have and there might be even more  but this is also how we are selling it because at [17:30] the end of the day you still need but you need  to upskill people you need to skill people for [17:35] different tasks and you need to do that. So that's  one element which we have as well. So but at the [17:42] end and now comes the target picture. This is how  it would look like in the future. So this is my [17:53] my picture in my head which somebody was able to  just bring on a on a slide. This is how I imagine [18:02] the agentic enterprise working. when you go really  into the autonomous stage and then you say hey [18:10] I have an orchestration layer and there  is not two three or four it's just one [18:18] and this orchestration layer is doing everything  what I need to do in order to run the agents [18:25] so it doesn't matter which functional area  it doesn't matter what the agent is doing [18:31] but this is exactly the picture which we have in  place that we say Hey, I can do all the crossf [18:39] functional use cases. I can create all the value  in if I want to have that. So you have basically [18:47] everything from a process point of view from a  functional point of view but you have all the [18:53] orchestration and you have this enterprise  foundation model in a sense in the background [18:59] that it's working and that you can have. So this  agentic layer is for us something where we say [19:05] if we can create that end to end and run  autonomously tasks complete tasks and do things [19:14] then we will have that and I think uh the  winners won't be those with the most AI features [19:21] I think the ones which are having this integrated  world will be as enterprise also successful in [19:29] doing that now this picture is lovely. I like it.  But uh to come there, it's a very hard work and [19:37] a very hard thing to do. But this is something  we are working on. We're working with companies [19:42] which give us the possibility to do that. And  now we need to look into how we can create that [19:49] more and more and build that up to get this  foundational model in place. So and then of course [19:56] um we come exactly to my picture at the beginning.  So the shift towards the AI enhanced processes [20:05] humans and digital workers are together and  now is of course also the time to do it because [20:13] important for us is if you're not  doing that if you're not on the journey [20:21] there is processes there is areas where  other companies will chime in because the [20:28] business domain knowledge is commoditized.  and they will take that over and they will [20:33] leave you just with some kind of core which is of  course not what we want. So we need to step in we [20:38] need to step up uh to do that that we are not  falling behind and it it is a journey for every [20:44] organization. It will be a journey which needs  to be done but uh I think if you create that [20:50] and if you are careful enough to bring that  all together you will be successful. Thank you.