• 亿邦公众号
    • APP下载

工业智能体告别“样板间”:吉利、杰克、美的三家制造业巨头的规模化落地真相

作者:虎嗅智库 文章来源:虎嗅智库服务

2026-06-29 10:41

作者:虎嗅智库 文章来源:虎嗅智库服务

2026-06-29 10:41

邦小白快读

EN
全文速览

本文分享了虎嗅智库举办的工业智能体进车间闭门会干货,披露了吉利、杰克、美的三家制造巨头AI工业智能体规模化落地的真实经验,核心结论是AI已经脱离样板间,从玩具变成可用工具,从试点进入量产阶段,有很多可落地的实操干货。

1. 核心落地逻辑:不要盲目先建大平台、数据中台,要场景驱动,只要场景明确、数据可用就可以落地,不要追求大而全,要让业务主导AI落地,不要做成面子工程。

2. 三家不同落地路径:吉利主打弱平台强智能体,补充传统软件的空白;杰克主打小场景切入,用小项目拿到真实提效成果再推广;美的建立了完整的规模化落地体系,已经落地上万个智能体。

3. 避坑指南:AI不会替代原有系统,走共生路线更稳妥,要提前明确ROI核算标准,不要对AI抱有不切实际的过高期待。

本文分享了头部制造品牌布局AI工业智能体的实操经验,对制造类品牌推进智能化转型、把握产业趋势、降本提效有很高参考价值。

1. 生产研发端价值:AI智能体可以渗透到缺陷管理、排产调度、工艺设计、工时测算、供应商管理等多个核心环节,能大幅压缩流程时间,比如杰克把单个零件工时测算从950秒降到26秒,美的把新机型工艺设计从2小时降到0.5小时,实打实降本提效。

2. 营销端价值:杰克用AI数字人陪练做营销培训,让经销商快速掌握产品技能,最终商机转化率提升30%,值得品牌借鉴。

3. 落地启示:不要贪大求全,先找痛点明确、数据充足的场景落地,要由业务部门一把手主导推进,把AI融入现有核心IT系统,让员工无感使用,同时要量化ROI,用降本增效的硬指标推进项目。

本文披露了工业AI智能体的最新发展动态,给制造类卖家指明了新的增长机会,也整理了可落地的实操方法和风险提示。

1. 市场机会:AI工业智能体已经从概念试点进入规模化落地阶段,在生产、管理、营销全链条都有明确的提效降本价值,是制造业接下来明确的增长赛道,提前布局就能拿到先发优势。

2. 可学习的落地方法:可以从小痛点小场景切入,先培养种子员工拿到真实成果再逐步扩散;AI不需要推翻原有IT系统,和现有系统共生即可,信息化程度低的企业甚至可以用智能体加即时通讯搭建轻量级流程,成本低落地快。

3. 风险提示:要避免对AI期望过高,需要正视AI幻觉、技术迭代快的问题,一定要提前明确ROI核算标准,避免项目做成没有实际价值的面子工程,所有投入都要对应可量化的降本增效成果。

本文分享了三家头部制造工厂AI智能体落地的真实经验,给工厂推进数字化智能化转型提供了非常务实的方向和启示。

1. 可落地的核心场景:AI智能体可以在多个高频生产管理场景创造价值,包括停线缺陷管理、智能排产、物料检索、设备维修、工时测算、工艺设计、供应商风险评估、精益管理等,能把核心环节的效率提升几倍到上百倍,大幅降低对老员工经验的依赖。

2. 转型启示:不要盲目跟风先建大平台、数据中台,要坚持场景驱动,只要场景明确、数据可用就可以先做起来;不要追求大而全,要由业务部门主导,把AI融入现有的PLM、MES等核心系统,让员工在日常使用中无感获得AI帮助;可以先培养种子员工树立典型,拿到成果再逐步推广。

3. 新商业机会:工厂可以借助AI能力升级,像杰克一样推出智能工厂交钥匙解决方案,开辟新的业务增长曲线。

本文披露了当前工业AI领域的行业发展趋势和制造企业的真实痛点,给制造业智能化服务商指明了业务方向和解决方案思路。

1. 行业发展趋势:当前工业AI智能体已经脱离早期炒概念的样板间阶段,进入真实规模化落地期,制造企业的需求从虚转实,不再关注泛泛的赋能,更关注真场景、真效果、可量化的ROI,市场需求已经进入爆发前期。

2. 客户核心痛点:大部分制造企业不清楚AI落地路径,怕做成面子工程,不知道怎么协调AI和原有系统的关系,不清楚怎么核算投入产出,大量企业落地多个智能体后缺乏统一的模型管理能力,知识沉淀没有形成闭环。

3. 解决方案参考:可以主打弱平台强智能体的服务方向,聚焦小场景解决真实痛点;主推AI和现有系统融合共生的方案,不需要客户推翻原有IT架构;要帮客户建立可量化的ROI核算体系,提供统一的算力和模型管理服务,助力客户完成知识治理,形成正向价值闭环。

本文披露了制造企业落地AI对平台的真实需求,也分享了头部平台的实践经验,给工业AI、工业互联网平台商的运营发展提供了明确参考。

1. 企业对平台的真实需求:企业不需要大而全的空泛平台,更需要聚焦具体场景的轻量化智能体服务,要求平台能够支撑多智能体的统一调度和模型管理,保障质量可控、成本最优、迭代高效、安全合规,同时要支持AI和企业现有各类IT系统顺畅对接融合。

2. 可借鉴的平台运营做法:头部企业美的的经验是搭建统一的算力平台和算法管理平台,支撑上万个智能体稳定运行,做到每个智能体的Token消耗、投入产出可观测可核算;吉利的经验是走弱平台强智能体路线,聚焦场景落地,不盲目扩张平台边界。

3. 风险规避:不要炒概念做样板间工程,要务实推进真场景落地,招商引导客户从小场景切入,避免客户盲目投入大平台建设,推动业务端主导对接,降低项目落地失败的风险。

本文披露了当前工业AI智能体产业的最新发展动向,梳理了头部企业的落地实践和暴露的问题,对产业研究有很高的一手资料价值。

1. 产业最新动向:AI工业智能体已经脱离早期试点的样板间阶段,正式进入规模化落地的实质发展期,成为制造业降本增效的真实可用工具,头部企业已经探索出三种成熟的落地路径,分别是吉利的弱平台强智能体路径、杰克的小场景快速落地路径、美的的全价值链规模化落地路径,三种路径都验证了AI的产业价值。

2. 产业现存新问题:当前AI工业落地普遍存在从业者期望过高、AI幻觉、技术迭代过快、原有系统过渡方案不清晰、ROI核算标准缺失、多模型管理难度大等问题,都是值得深入研究的方向。

3. 新的研究命题:当前行业已经形成场景驱动、业务主导、数据先行、与现有系统共生的全新落地逻辑,和传统制造业信息化路径完全不同,同时AI倒逼企业完成知识治理形成价值闭环,都是产业研究的新方向。

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

This article shares key insights from a closed-door workshop on industrial AI agents for factory floors hosted by Huxiu Think Tank. It reveals real-world scaling experiences from three manufacturing giants: Geely, Jack, and Midea. The core conclusion is that AI has moved beyond prototype demos: it is no longer just a "proof of concept" toy, but a production-ready usable tool that has progressed from pilot testing to mass deployment, with plenty of actionable, implementable takeaways.

1. Core deployment logic: Do not rush to build large platforms or data meshes blindly. Instead, prioritize scenario-driven implementation: roll out AI as long as the use case is clear and data is available. Avoid pursuing overly broad, one-size-fits-all projects; let business teams lead AI deployment rather than building it as a superficial vanity project.

2. Three distinct scaling paths: Geely focuses on "weak platform, strong agents" to fill gaps in traditional enterprise software. Jack starts with small, targeted use cases, delivers measurable efficiency gains through small pilot projects, then scales up from there. Midea has built a complete, scalable deployment system with over 10,000 AI agents already in operation.

3. Common pitfalls to avoid: AI will not replace existing systems. A coexistence strategy is far more reliable. Define clear ROI metrics in advance, and do not hold unrealistic, overly high expectations for AI.

This article shares hands-on experience from leading manufacturing brands building out industrial AI agents, offering valuable insights for manufacturing brands pursuing intelligent transformation, tracking industry trends, and cutting costs while boosting efficiency.

1. Value for production and R&D: AI agents can be integrated into multiple core workflows including defect management, production scheduling, process design, labor hour calculation, and supplier management, delivering significant time compression. For example, Jack cut individual part labor hour calculation from 950 seconds to 26 seconds, while Midea reduced new model process design from 2 hours to just 0.5 hours, delivering tangible cost reduction and efficiency gains.

2. Value for marketing: Jack uses AI digital humans for interactive sales training, helping distributors master product knowledge quickly and ultimately boosting lead conversion rates by 30%, a tactic worth replicating for other brands.

3. Key takeaways for deployment: Avoid overambitious, overly broad projects. Start with clearly defined pain points and use cases with sufficient available data. Have the head of the relevant business department lead the initiative, integrate AI into existing core IT systems to enable seamless, invisible use for employees, and quantify ROI to advance projects with hard cost-efficiency metrics.

This article outlines the latest developments in industrial AI agents, identifies new growth opportunities for manufacturing-focused sellers, and organizes actionable implementation methods along with risk warnings.

1. Market opportunity: Industrial AI agents have moved beyond conceptual pilots and entered the large-scale deployment phase. They deliver clear cost reduction and efficiency improvements across the entire value chain of production, management, and marketing, making it a clear high-growth track for the manufacturing sector. Early adoption will deliver first-mover advantage.

2. Actionable deployment lessons: You can start with small pain points and targeted use cases, train early "seed employees" to deliver tangible results, then expand gradually. AI does not require replacing existing IT infrastructure; it can work alongside current systems. Even businesses with low digital maturity can build lightweight workflows by combining AI agents with instant messaging tools, enabling low-cost, fast deployment.

3. Risk warnings: Avoid setting overly high expectations for AI; acknowledge existing challenges including AI hallucinations and rapid technology iteration. Always define clear ROI calculation standards in advance to avoid turning projects into valueless vanity projects, and ensure all investments tie to quantifiable cost and efficiency gains.

This article shares on-the-ground AI agent deployment experience from three leading manufacturing factories, offering practical direction and insights for factories pursuing digital and intelligent transformation.

1. Actionable high-impact use cases: AI agents create value across many high-frequency production and management scenarios, including downtime and defect management, intelligent scheduling, material search, equipment maintenance, labor hour calculation, process design, supplier risk assessment, and lean management. They can boost core workflow efficiency by multiples to 100x, and significantly reduce reliance on veteran employees' tacit knowledge.

2. Transformation takeaways: Do not blindly follow the trend of building large platforms or data meshes first. Stick to a scenario-driven approach: you can get started as long as your use case is clear and data is available. Avoid pursuing overly broad solutions; let business teams lead integration, embed AI into existing core systems such as PLM and MES so employees get AI support seamlessly in their daily work. Start by training seed employees to build proven use cases, then scale incrementally.

3. New business opportunities: Factories can leverage AI capabilities for upgrading, just as Jack did by launching turnkey smart factory solutions to open up new growth curves.

This article reveals current industry trends in industrial AI and the real pain points of manufacturing enterprises, outlining clear business directions and solution frameworks for intelligent manufacturing service providers.

1. Industry trends: Industrial AI agents have moved beyond the early concept-driven prototype stage and entered a period of real large-scale deployment. Manufacturing demand has shifted from abstract to concrete: businesses no longer care about vague "empowerment" and instead prioritize real use cases, proven results, and quantifiable ROI. Market demand is now on the cusp of explosive growth.

2. Core customer pain points: Most manufacturing enterprises lack clarity on AI deployment paths, fear building vanity projects, struggle to coordinate the relationship between AI and existing systems, and lack clear frameworks for calculating return on investment. After deploying multiple AI agents, many enterprises still lack unified model management capabilities, and fail to form closed-loop knowledge沉淀.

3. Reference solution frameworks: Focus on a "weak platform, strong agents" service strategy, concentrating on solving real pain points in small targeted use cases. Prioritize solutions that integrate and coexist with existing systems, so clients do not need to replace their current IT architecture. Help clients establish a quantifiable ROI calculation framework, provide unified computing power and model management services, support clients in completing knowledge governance, and build a positive closed value loop.

This article reveals manufacturers' real demand for platforms when deploying AI, shares practical experience from leading platforms, and offers clear reference for the operation and development of industrial AI and industrial internet platform providers.

1. Manufacturers' real demand for platforms: Enterprises do not need broad, vague all-in-one platforms. They prioritize lightweight AI agent services focused on specific use cases, and require platforms to support unified scheduling and model management for multiple agents, to ensure controllable quality, optimal cost, efficient iteration, and security and compliance. They also require smooth integration between AI and their existing various IT systems.

2. Actionable platform operation lessons: Leading manufacturer Midea's experience is to build unified computing power and algorithm management platforms to support the stable operation of over 10,000 AI agents, with measurable and trackable token consumption and ROI for each individual agent. Geely's approach follows the "weak platform, strong agents" path, focusing on use case deployment rather than blindly expanding platform scope.

3. Risk mitigation: Avoid hyping concepts or building only prototype demo projects. Push for practical implementation in real use cases, guide clients to start with small scenarios to avoid wasteful overinvestment in large platforms, and have business teams lead collaboration to reduce the risk of deployment failure.

This article reveals the latest industry developments in the industrial AI agent sector, organizes deployment practices from leading enterprises and the challenges that have emerged, providing valuable first-hand data for industry research.

1. Latest industry developments: Industrial AI agents have moved beyond the early prototype pilot stage and formally entered the substantive development phase of large-scale deployment, becoming a real usable tool for manufacturing cost reduction and efficiency improvement. Leading enterprises have developed three mature deployment paths: Geely's "weak platform, strong agents" model, Jack's small-scenario fast deployment model, and Midea's full-value-chain large-scale deployment model, all of which have verified the industrial value of AI.

2. Current emerging industry challenges: Common pain points in industrial AI deployment include overly high industry expectations, AI hallucinations, rapid technology iteration, unclear transition plans for legacy systems, lack of standardized ROI measurement, and difficulty managing multiple large models. All of these represent areas worthy of in-depth research.

3. New research propositions: The industry has now formed a completely new deployment logic: scenario-driven, business-led, data-first, and coexistence with existing systems, which differs entirely from the traditional manufacturing informatization path. Additionally, AI is forcing enterprises to complete knowledge governance and form closed value loops. Both are new directions for industrial research.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

大模型聊到今年,说句实在话,我和很多制造、能源行业的数字化老兵一样,多多少少都有些概念疲劳了。

AI到底能不能在工厂里帮企业赚到钱?怎么用才不会变成一把手工程的“面子花瓶”?这是我们当下最想知道的。

带着这层极度务实的焦虑,虎嗅智库上周搞了这场“Agent进车间主题闭门会”。

制造业的数字化分享我们办过不少,但把“AI智能体”和“车间”这两个词放一块儿,总让人觉得透着股“悬浮感”——甚至连部分参会的同行,一开始也并不看好这种跨界实践。

所以会前,我们和吉利集团广域铭岛的AI工业负责人张兴、杰克科技流程与IT总监陈恭伟、美的集团美云智数AIGC解决方案专家张勇这三位分享嘉宾,反复对齐:一定要讲真场景真案例,哪怕聊聊踩过的坑。同时,我们严格审核,最后进入线上会议的,都是制造和能源行业负责技术落地的同行,保障交流质量。

结果?整整两个小时的会议,全程没有尿点,200多位观众直呼收获满满。那一刻我强烈地意识到:AI智能体其实早就扎进工业深水区,在车间里打响真正的硬仗了。

张兴:别急着建平台,先找到那个闭环

第一位上来分享的是广域铭岛的AI负责人张兴。作为吉利在10年前就孵化出的工业互联网“国家队”,他们现在正全速向AI转型。张兴基于丰富的集团内外案例,深度拆解了传统软件与AI智能体的本质差异,并提出了“弱平台,强智能体”的核心落地理念。

在张兴看来,传统软件和AI智能体根本不是一回事。传统软件是“流程驱动”,立项先画流程图,每个节点输入输出理清楚;而智能体是“数据驱动”,第一句话问的是——你有什么数据?能从里面挖出什么洞察?

这个差异直接决定了落地路径的不同。他分享了几个已经跑起来的场景:

停线和缺陷管理——这是制造业的“老冤家”。以前质量问题靠人盯,缺陷报告躺在系统里吃灰。现在智能体做的是完整闭环:识别问题、根因分析、对策推荐、知识沉淀。更狠的是解决问题背后的问题——通过DPV(单车缺陷数)分析,找出那些反复出现的管理漏洞。“一线工人解决直接问题,智能体帮管理层看到问题背后的问题。”

智能调度——计划员的排产从来不是标准化流程。缺料了怎么办?订单变更了怎么调?这些靠经验意会的东西,以前APS系统根本抓不住。现在智能体像陪练一样跟着计划员,记录他在不同场景下的决策逻辑,慢慢提炼出行为偏好。“就像你用龙虾,越纠正它越懂你。”

信息过载——采购人员跑完MRP,真正的痛苦才开始:哪些物料缺了?为什么缺?建议下单数和历史差异大不大?以前一个采购员一天处理20个物料就焦头烂额,现在智能体先做全局风险扫描,再把"有异常的5个物料"推到人面前确认。“以前软件跑完就完了,现在软件跑完,智能体才开始干活。”

张兴反复强调一个观点:智能体不是替代传统软件,而是填补传统软件的空白。“你的APS都没用好,指望智能体帮你做好排产?路都没有,怎么跑得快?”

现场有人问起“数据中台和智能体哪个先建”,张兴的回答很干脆:“场景驱动。两个条件:场景能不能明确?数据有没有?都有就干,没有就先别折腾平台。

陈恭伟:缝制设备龙头的“土办法”与真效果

接下来,杰克科技的陈恭伟带来了更多“车间层”的务实案例分享。

这家A股上市的缝制设备龙头,今年6.16刚在越南办了场发布会,推整套智能工厂的“交钥匙”方案——智链大师。但陈恭伟没讲这些,他讲的是工程师找物料怎么找、新员工修设备怎么修、工艺工程师怎么算工时——全是工厂里天天发生的小事儿,也是高价值场景。

物料检索——以前工程师在PLM系统里搜“台板”,出来几百个型号,非专业人士根本找不到匹配的。现在对着AI助手说一句“适合M9机型的台板”,秒级定位,直接跳转到图纸和工艺包。“效率提升5到10倍,不是那种预计提升,是真金白银的计时。”

设备维修——老师傅的经验是最稀缺的资源。杰克把维修手册、历史案例建成知识图谱,集成到SCADA报修系统。新员工报修后,AI直接推荐“可能原因+历史解决方案”,不用再翻手册、不用再喊老师傅。

工时测算——这是全场最为惊艳的数据之一。传统MTM(方法时间测量)需要资深精益工程师,一个零件赋码要950秒,一台机器500多个零件,得干16天。现在AI学习BOM、工艺、工时数据后自动生成装配步骤,再匹配标准工时库,26秒搞定一个零件,整机一天就能赋完码。“AI生成的方案,相当于5年经验的精益工程师水平。”

陈恭伟还透露了一个内部数据:数字人陪练在营销场景落地后,商机转化率提升了30%。“以前培训是学完了不知道会不会用,现在是先学再练,雷达图直接标出薄弱环节。”

但他也坦承了AI落地的坑:期望太高、幻觉问题、AI技术的更新迭代太快。杰克的解法是“松土、树典型、找痛点”——先让“金种子”员工尝到甜头,再慢慢扩散;同时,坚决不能求大而全,必须由业务部门一把手主导,将AI技术深度融进PLM、MES等核心IT系统,让用户在系统便利性中无感感知AI的价值。

张勇:美的的180个智能体与模型管理

最后一位出场的张勇老师,作为智能制造领域的绝对标杆,美的集团在AI领域的探索展现出了大厂在算力、机制与全价值链覆盖上的压倒性深度。

美的的AI发展分了三个阶段:2023年的“工具阶段”(IT主导,价值探索阶段)、2024年的“业务融合阶段”(业务牵头,战略小组自上而下推动)、现在的“模式变革阶段”(端到端拉通,改变业务模式)。

“最关键的转变是让业务主导,而不仅仅是IT。”张勇说。美的的推法是"5板斧":战略制定、预算、KPI、运营、变革。年初给各事业部下"降本增效"的硬指标,财务引导业务方立项时优先选AI项目——"你不立AI项目,又达不到指标,就说不过去了。"

目前美的官方运营的智能体有180多个,员工"手搓"的超过2万多个。支撑这些智能体调度的算力及模型管理都集中在统一的算力平台和算法管理平台上,张勇特别强调"模型管理"的重要性——"模型很多,需要统一的管理调度,保障质量可控、成本最优、迭代高效以及安全合规。"

他分享的工厂场景同样扎实:

DMS精益管理——借助DMS智能体,将日常管理、异常处置流程标准化、自动化,缩短会议时间,提升异常闭环质量,同时可以将优秀管理者的经验转化为系统规则,降低对个人经验的依赖,加速新人培养。"减少会议次数,缩短开会时间,班组长的时间很宝贵,流水线一直在跑。"

工艺设计——以前新机型从0到1设计工艺要2小时,现在AI基于历史工艺库匹配相似方案,设计师微调即可,降到0.5小时。

供应商风险——四大维度的风险评估,从"人工评估"变成"AI全量实时评估"。难点在于需要供应商把产能数据接进来,"美的是链主,能拉通供应商配合,但不是所有企业都能干这个事"——这话实在。

张勇特别强调了"知识治理"的重要性。"以前做知识库,就是把文档存起来,好不好不知道。现在AI用上了,回答错了马上反馈,知识Owner就得去修正。AI倒逼知识治理,形成了价值闭环。"

现场那些热烈交流

由于整场闭门会分享的内容极具实操性,三位嘉宾分享结束后的答疑环节被各种问题塞满。我们整理了几个最具代表性的:

Q:AI智能体和原有系统怎么过渡?

张兴的答案是“共生而非替代”。信息化程度高的企业,智能体依附于现有系统做提效;信息化程度低的,反而可能跳过传统软件,用“智能体+即时通讯”直接搭轻量级流程。“有些场景,ERP没上全,但智能体已经能跑起来了。”

Q:ROI怎么算?

张勇展示了美的的硬核做法:所有Token消耗可观测,每个部门、每个智能体的调用量实时统计;立项时财务、人力、IT三方确认提效标准(比如翻译场景,原来几天vs现在几分钟);结项时三方再核算。“不是定性地说'效率提升了',是真金白银的工时和成本。”

我们(虎嗅智库)的一点感受

办这场会之前,我们担心Agent智能体在工业场景还是个伪命题。但三位嘉宾的分享让我们确认了一件事:AI在车间里,已经开始从“玩具”变成“工具”,从“试点”走向“量产”。

吉利强调“弱平台、强智能体”,杰克科技证明“小场景也能出大效果”,美的展示了“规模化落地的方法论”。三家企业路径不同,但共识高度一致:找场景、治数据、让业务主导、容忍AI慢慢成长。

会后交流群里的讨论持续不断。有参会者说:“终于听到不说赋能只说怎么干的会了。”另一位补充:“原来对Agent是什么、怎么用比较茫然,听完心里踏实多了。”

这或许就是闭门会的价值——不是听成功案例有多光鲜,而是知道哪些坑其实可以避开,哪些焦虑其实大可不必。

注:文/虎嗅智库,文章来源:虎嗅智库服务(公众号ID:huxiuzhikufuwu),本文为作者独立观点,不代表亿邦动力立场。

【版权提示】亿邦动力倡导尊重与保护知识产权。未经许可,任何人不得复制、转载、或以其他方式使用本网站的内容。如发现本站文章存在版权问题,烦请提供版权疑问、身份证明、版权证明、联系方式等发邮件至run@ebrun.com,我们将及时沟通与处理。

已点赞

+1

{{ praise_text }}

好文章,值得被更多人看到

FAQ回顾

工业智能体在制造业落地有哪些可参考的实践经验?

吉利提出“弱平台强智能体”理念,优先选择有明确场景、可用数据的领域落地,填补传统软件空白;杰克科技聚焦物料检索、设备维修等小场景提效,工时测算效率较传统方式提升超36倍;美的已有180个官方运营智能体,通过业务主导、统一算力管理实现规模化落地。

制造企业落地AI智能体如何计算投入产出比?

可参考美的的测算方式,全链路观测每个部门、每个智能体的Token消耗与调用量,立项时由财务、人力、IT三方共同确认提效标准,结项时再共同核算实际节省的工时与成本,实现可量化的收益评估。

工业智能体和企业原有IT系统是替代关系吗?

工业智能体并非替代传统软件,而是填补传统软件的空白,二者为共生关系。信息化程度高的企业,智能体可依附现有系统提效;信息化程度较低的企业,甚至可跳过部分传统软件,用智能体搭配即时通讯搭建轻量级流程。

制造企业落地工业智能体常见的问题有哪些?

常见问题包括对AI期望过高、AI存在幻觉问题、AI技术迭代速度快等,可参考杰克科技的解法:先培养金种子员工打造典型案例,聚焦真实业务痛点,由业务部门主导将AI融入核心IT系统,避免求大而全。

广告

下一篇

跨境电商资源对接

《全球跨境电商服务商资源手册》第二期手册现已发布,触达更多资源,了解跨境服务商信息,欢迎点击链接下载。

了解详情

有人投诉就“限单”?商品品质差要被抖音电商严惩了

挂羊头卖狗肉,只要是描述与售卖不一致(商品名称、规格、成分、材质、产地、年份等),或者做工瑕疵、商品破损/污损/变形、商品变质/异物等等,在抖音电商都会被认定为“商品品质差”的违规行为。