• 亿邦公众号
    • APP下载

当Agent大规模落地 IT部门将成为下一个HR部门??【线下闭门会】

作者:虎嗅智库 文章来源:虎嗅智库服务

2026-06-15 10:22

作者:虎嗅智库 文章来源:虎嗅智库服务

2026-06-15 10:22

邦小白快读

EN
全文速览

本文是虎嗅智库主办的消费零售产业AI Agent落地闭门会的内容整理,披露了企业落地AI的真实卡点、共识和可落地方法,核心干货如下:

1. 当前企业AI落地存在普遍误区,很多企业做了上百上千项单点AI应用,局部降本提效效果明显,但因为没有触达组织调整、人员精简、流程重构三个层面,无法带来企业级的改变,得不到企业和管理层的认可,属于无效落地。

2. 业内总结出了更高效的AI落地产出模式,替代过去沟通成本高、交付慢的传统IT模式,也就是采用“玩家平权”加“双人结对小组”模式,让懂业务的需求方和懂AI的数字化BP结对,由BP做AI能力翻译,减少Token浪费,快速完成交付。

3. 明确了IT部门的转型方向,同时披露了大模型领域解决记忆混乱、精度下降问题的最新技术方向。

本次闭门会汇集了多个消费零售头部品牌的AI操盘手,讨论了品牌落地AI的真实问题,对品牌布局AI有较高参考价值,核心干货如下:

1. 落地误区提示:不少头部品牌已经做了数百项AI应用,覆盖ChatBI、智能客服、AI配方研发、营销短视频混剪等场景,营销成本降低了三四倍,但因为只是单点工具提效,没有推动组织、人员、工作流程的深层变革,老板和企业都没有感知,属于无效落地。

2. 可借鉴的落地模式:采用“双人结对小组”模式,让懂业务的营销、电商人员和懂AI的数字化BP结对协作,快速把AI推到业务前端,解决传统IT交付排期慢、跟不上业务变化的问题,同时减少AI调用的无效浪费,提升响应效率。

3. 明确了品牌AI的未来发展方向,未来AI会向Super Agent加超级基座的架构演进,品牌需要推动IT部门转型,提前布局AI基建和治理体系。

本次闭门会披露了AI落地消费零售领域的真实现状,给出了卖家布局AI的风险提示和可落地经验,核心干货如下:

1. 风险提示:当前很多卖家布局AI容易陷入单点工具误区,虽然局部降本提效效果明显,但是没有带来企业层面的变革,无法获得管理层和企业层面的资源支持,最终属于无效落地,做了很多工作也无法证明价值。

2. 可借鉴的落地方法:可以放弃过去业务提需求、产品画原型、技术排期开发的传统模式,改用“双人结对小组”模式,让懂业务的人员和懂AI的BP结对协作,快速交付AI应用,跟上业务变化的速度,还能减少Token浪费,提升AI响应效率。

3. 机会提示:当前AI Agent已经进入规模化落地阶段,业务端可以自主借助AI框架搭建应用,只要找准落地方法,就能快速借助AI实现降本增长,而IT部门转型为AI基建治理者后,也能给业务AI落地提供更好的支撑。

本次闭门会讨论的企业AI落地逻辑,对工厂推进数字化和AI转型有很强的启示意义,核心干货如下:

1. 转型误区警示:工厂推进数字化AI转型,很容易陷入单点工具的误区,也就是只开发服务单个岗位的提效工具,没有推动组织调整、人员精简和核心生产流程的重构,最终无法获得企业层面的认可,投入很多资源也无法体现价值,属于无效转型。

2. 落地路径启示:可以借鉴消费零售行业验证的“双人结对”模式,让懂生产、设计、供应链的业务人员和懂AI的技术BP结对协作,快速把AI落地到业务环节,解决传统IT交付周期长,跟不上业务变化的老问题。

3. 技术方向启示:未来AI Agent会走向按需自动组装、自迭代优化的方向,工厂布局AI不需要一开始就搭建固定的庞大系统,可以逐步搭建灵活的AI应用体系,同时推动IT部门转型为AI基建和治理者,支撑业务端的AI创新。

本次闭门会披露了当前企业落地AI Agent的真实痛点和行业发展趋势,对AI服务商优化解决方案有很高参考价值,核心干货如下:

1. 客户核心痛点梳理:当前大量企业客户已经完成了AI试点,落地了很多单点AI应用,局部降本效果明显,但无法实现企业级的价值落地,得不到管理层认可;同时传统IT交付模式太慢,跟不上业务进化速度,技术层面还存在大模型上下文污染、记忆混乱、大规模智能体运行效率下降的痛点。

2. 客户需求方向:当前企业不再满足于单一的成功案例分享,更需要可迁移的落地解法,需要能支撑“双人结对”落地模式的工具,也需要具备记忆隔离、自反思迭代能力的技术底座,解决现有大模型的技术痛点。

3. 行业发展趋势:未来企业AI会从人机协作走向智能体自协同,主流架构会变成Super Agent加超级基座,固定形态的Agent会被按需动态组装的Agent取代,服务商需要顺着这个方向迭代产品。

本次闭门会传递了当前产业端对AI平台和行业服务平台的真实需求,也给出了平台运营转型的参考方向,核心干货如下:

1. 产业端的真实需求:当前企业落地AI已经度过了试点阶段,不再满足于看头部企业的成功案例,更关注落地方法的可迁移性,需要聚集产业一线共创真实卡点解法的交流场景,单一的成功分享参考价值已经很低。

2. 行业交流平台的转型方向:可以借鉴虎嗅智库的转型经验,从单纯的内容观察者转向行业判断与共识形成平台,聚焦垂直细分的落地主题,定期举办线下闭门会、线上圆桌,组织产业一线拆解真实问题,后期还可以通过评选优质落地企业沉淀行业资源,打造自身影响力。

3. AI技术平台的发展方向:企业需要能支撑业务端自主搭建AI应用的平台,需要平台具备记忆隔离、自反思迭代的技术能力,支撑Super Agent的架构需求,同时需要平台赋能企业的数字化BP,帮助企业落地新的AI协作模式。

本文记录了国内消费零售产业AI Agent落地一线的最新讨论,呈现了产业落地的最新动向和真实问题,有很高的一手研究价值,核心内容如下:

1. 产业落地出现的新问题:当前AI落地已经从0到1的试点阶段进入规模化落地阶段,出现了规模化单点落地但无法创造企业级价值的新问题,同时AI技术前置让传统IT部门的定位受到挑战,“IT部门是否会消失”成为产业界普遍焦虑的新问题。

2. 产业和技术的新动向:落地模式上,诞生了“双人结对小组”的新协作模式,替代传统的需求交付模式;技术上,行业正在探索记忆隔离、反思层自迭代的技术方向,解决大模型上下文污染、精度下降的痛点,未来Agent会走向动态按需组装的新形态;组织上,明确了IT部门从需求交付者转向AI基建与治理者的演进方向。

3. 行业共识的新变化:当前产业界已经从案例崇拜转向解法共创,更关注落地方法的可迁移性,这也是产业研究需要跟进的新方向。

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 summarizes content from a closed-door AI Agent implementation workshop hosted by Huxiu Institute for the consumer and retail industry. It reveals real-world roadblocks, industry consensus, and actionable methods for enterprise AI adoption, with key takeaways below:

1. There is a widespread misconception in current enterprise AI implementation: many companies have deployed hundreds or thousands of point AI solutions that deliver clear local cost reduction and efficiency gains. However, these projects have not touched organizational restructuring, headcount optimization, or process reengineering, so they fail to drive enterprise-level transformation and gain recognition from leadership, classifying them as "ineffective implementation".

2. The industry has developed a more efficient AI delivery model that replaces the traditional IT approach, which suffers from high communication costs and slow delivery. The new model combines "equal participation across all stakeholders" with "paired two-person teams": business stakeholders with domain expertise pair with digital business partners (BP) who understand AI. The BP translates business requirements into AI-ready language, cutting unnecessary token waste and enabling rapid delivery.

3. The discussion clarified the transformation direction for IT departments, and also disclosed the latest technical approaches to solving large language model problems including memory confusion and declining output accuracy.

This closed-door workshop brought together AI leaders from top consumer and retail brands to discuss real-world challenges of AI adoption for brands, offering high-value actionable insights. Key takeaways are as follows:

1. Common implementation pitfalls: Many leading brands have already deployed hundreds of AI applications across use cases including ChatBI, intelligent customer service, AI-powered formula R&D, and marketing short-video remixing, cutting marketing costs by 3-4x. However, these are all point solutions that only improve efficiency at the local level, without driving deep changes to organization, headcount, and work processes. As a result, neither executive leadership nor the broader enterprise feels the impact, making this ineffective AI implementation.

2. A replicable implementation model: Adopt the "paired two-person team" model, where business-focused marketing and e-commerce staff pair and collaborate with AI-savvy digital BPs. This approach quickly pushes AI capabilities to the frontline of business, solves the problem of slow delivery queues and inability to keep up with business changes that plagues traditional IT delivery, and also reduces unnecessary AI token waste while improving response efficiency.

3. The future development direction for brand AI is clarified: AI will evolve toward an architecture of Super Agent plus a super foundational base. Brands need to drive the transformation of their IT departments, and build out AI infrastructure and governance systems in advance.

This closed-door workshop disclosed the real current state of AI adoption in consumer and retail, shared risk alerts and actionable experience for sellers looking to adopt AI. Key takeaways are below:

1. Risk alert: Many sellers currently fall into the pit of only adopting point AI tools. While these deliver clear local cost reduction and efficiency gains, they do not drive enterprise-level transformation, so sellers struggle to secure buy-in and resources from management, and ultimately cannot demonstrate the value of their work, making it ineffective implementation.

2. Actionable replicable methods: Sellers can abandon the traditional model where business teams submit requirements, product teams build prototypes, and engineering teams schedule development, and instead adopt the "paired two-person team" model. When business-savvy staff pair and collaborate with AI-knowledgeable BPs, they can deliver AI applications quickly enough to keep up with the pace of business change, cut token waste, and improve AI response efficiency.

3. Opportunity note: AI Agent has now entered the stage of large-scale implementation. Frontline business teams can build AI applications independently using existing AI frameworks. With the right implementation approach, sellers can quickly leverage AI to cut costs and drive growth. Once IT departments transform into AI infrastructure and governance managers, they can also provide better support for business-side AI implementation.

The logic of enterprise AI implementation discussed at this closed-door workshop offers strong instructive insights for factories advancing digital and AI transformation. Key takeaways are as follows:

1. Common transformation pitfalls: When advancing digital and AI transformation, factories often fall into the trap of only building point tools, building efficiency tools that serve only single roles, without driving organizational adjustment, headcount optimization and core production process reengineering. Ultimately, these projects cannot win enterprise-level recognition, fail to demonstrate value despite heavy resource investment, and qualify as ineffective transformation.

2. Implications for implementation paths: Factories can adopt the "paired two-person team" model validated in the consumer and retail industry, pairing business staff who understand production, design and supply chain with AI-savvy technical BPs to quickly implement AI into business workflows. This solves the long-standing problems of long traditional IT delivery cycles and inability to keep up with business changes.

3. Implications for technical direction: Future AI Agents will move toward on-demand automatic assembly and self-iterative optimization. Factories do not need to build a large, fixed system from the start when adopting AI; they can gradually build a flexible AI application system, while pushing IT departments to transform into AI infrastructure and governance managers to support AI innovation on the business side.

This closed-door workshop disclosed real-world pain points and industry development trends for current enterprise AI Agent implementation, offering high-value insights for AI service providers looking to optimize their solutions. Key takeaways are below:

1. Core customer pain points: A large number of enterprise customers have already completed AI pilots and deployed many point AI applications that deliver clear local cost reduction. However, they have failed to deliver enterprise-level value and win management recognition. Meanwhile, traditional IT delivery models are too slow to keep up with the pace of business evolution. On the technical side, large language models still suffer from context contamination, memory confusion, and declining operational efficiency when running large-scale agent networks.

2. Direction of customer demand: Enterprises are no longer satisfied with single successful case studies; they now demand replicable implementation methodologies, tools that support the "paired two-person team" implementation model, and a technical foundation with memory isolation and self-reflection iteration capabilities to solve the existing technical pain points of large language models.

3. Industry development trends: Enterprise AI will evolve from human-machine collaboration to agent-to-agent self-collaboration, with the mainstream architecture shifting to Super Agent plus a super foundational base. Fixed-form agents will be replaced by dynamically assembled on-demand agents, and service providers need to iterate their product roadmaps to align with this direction.

This closed-door workshop conveyed the real demand from industry players for AI platforms and industry service platforms, and outlined reference directions for platform operation transformation. Key takeaways are as follows:

1. Real demand from industry players: Enterprise AI adoption has now passed the pilot stage. Companies are no longer satisfied with just reviewing successful cases from leading firms; they now prioritize the replicability of implementation methods, and need communication spaces where frontline industry players can collaborate to develop solutions for real implementation roadblocks. Single case studies now offer very limited reference value.

2. Transformation direction for industry exchange platforms: Platforms can learn from the transformation experience of Huxiu Institute, shifting from being purely content observers to platforms for building industry consensus and analysis. They should focus on niche, implementation-focused themes, hold regular offline closed-door workshops and online roundtables, organize frontline industry players to dissect real problems, and later build influence and aggregate industry resources by recognizing high-quality AI implementation cases.

3. Development direction for AI technology platforms: Enterprises need platforms that support business teams to independently build AI applications, with technical capabilities including memory isolation and self-reflection iteration to support the Super Agent architecture. Platforms also need to empower enterprises' digital BPs to help companies implement the new AI collaboration model.

This article records the latest frontline discussion on AI Agent implementation in China's consumer and retail industry, presenting the latest trends and real problems of industry adoption, with high value as primary research. Core content is as follows:

1. New problems emerging in industry adoption: AI adoption has moved from the 0-to-1 pilot stage to large-scale implementation. A new problem has emerged: players are rolling out point solutions at scale but failing to create enterprise-level value. Meanwhile, the advancement of AI technology has challenged the traditional positioning of IT departments, and "will IT departments disappear?" has become a common source of anxiety across the industry.

2. New industry and technology trends: In terms of implementation models, a new collaborative model of "paired two-person teams" has emerged to replace the traditional requirement-delivery model. On the technical side, the industry is exploring memory isolation and reflection layer self-iteration to solve large language model pain points including context contamination and declining accuracy. In the future, agents will take the new form of on-demand dynamic assembly. On the organizational side, the evolution path for IT departments has been clarified: they will transition from requirement deliverers to AI infrastructure and governance managers.

3. New shifts in industry consensus: The industry has shifted from case worship to collaborative solution building, and now prioritizes the replicability of implementation methods. This is also the new direction that industry research needs to follow.

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.

十天前,在北京万科时代中心二层的一间会议室里,发生了一场充满焦虑、对抗、自我怀疑的真实对话。

“未来数字化部门可能没有传统意义上的产品经理和纯开发角色了,传统IT交付模式的排期太慢,根本跟不上业务的进化。”

“当业务自己能用AI搞定一切,那还要IT部门干什么?IT部门拿什么证明存在感?”

“AI是一个大大的海啸,每个人的山头都会被吞掉的。”

这场虎嗅智库主办的——消费零售产业AI Agent落地闭门会,从案例深度复盘开始,到全员拆解各自AI落地场景与业务卡点,共持续了三个多小时。

与会嘉宾包括自然堂集团大数据中心总经理罗予晋、蒙牛集团数据中台高级总监邢昊、丽人丽妆AI战略负责人李京京、华润三九AI研究院执行院长李晓真、德胧酒店集团CTO孟令航、龙湖集团千丁数科CTO刘怀洋、杭州联华华商CIO沈铮、金山办公高端制造事业部总经理于叶舟、飞书消费综合行业战略客户总监滕乃宗、上品集团CTO张诚、红熊AI记忆科学首席产品官揭志敏、AI营销专家贾梦,以及北大国发院教授周宏骐等。

作为发起人,虎嗅智库原本设想的议程很标准:嘉宾依次分享Agent落地成果、开源智能体带来的增长机会、业务提效的实施路径。冷静,专业,且充满希望。却万万没想到刚开场,嘉宾不约而同的“只报忧、不报喜”。

整场会议结束后,我的笔记本上没有最佳实践的漂亮框架,只记满了数字、问题和冲突。以下是我从现场带回来的几点观察。

735项应用落地,为何老板依然“没感觉”

会议的第一个高潮,来自一位美妆品牌数字化负责人的复盘。

团队从2023年开始做AI,两年做了735项应用,ChatBI、智能客服、AI配方师、短视频混剪,营销成本降了三四倍,这在局部看是非常漂亮的成绩。“但2025年底复盘时才发现,企业和老板对我们做的这些事,几乎一点感觉都没有。”

我注意到台下几位CIO下意识点了点头,卡点恰恰就在这里:当上千个AI的小功能、新特征被零散地分布在不同的岗位上,仅仅作为个人的便利提效工具时,它并没有触发企业级层面的真实改变。

他接着抛出一个判断:“企业对新势力的感知和改变,只会在三个地方发生:第一,有没有新组织或者旧组织的调整;第二,有没有人员的精简与组织瘦身;第三,工作流、流程和SOP有没有发生根本性变化。如果这三点都没有,做再多的单点AI工具,在企业治理层面都是无效落地。”

现场随即进入了讨论,大家对比了两种模式:过去的模式是业务提需求、BP接单、产品经理画原型、技术排期开发,沟通成本极高,瓶颈极多。而现在更有效的解法是是“玩家平权”与“双人结对小组”——让懂业务的需求方与懂AI对话、能进行结构化输入的数字化BP结对协作,由BP作为"AI能力翻译者",用最少Token、最快速度完成交付,把AI直接推到营销与电商的最前端。

结对,是为了让AI真正"听得懂"业务。当前的AI尚未聪明到能完全替代人类,单靠普通的业务人员去提问,往往会造成巨大的Token浪费、响应低效甚至南辕北辙。

IT部门的退路与进化:从“需求交付者”到“AI基建与治理者”

当技术能力前置、业务部门开始通过无代码或Agent框架“万马奔腾”地自己搞应用时,IT部门是不是要失业了?这是现场蔓延的另一个核心焦虑。

在全程见证了各方争论之后,周宏骐教授的观察一锤定音。他指出,从他今年调研的几十家AI企业和海外硅谷的最新趋势来看,企业AI正在经历从人机协作(H2A)到智能体自进化与协同的演进。未来的趋势是Super Agent(主代理管数十个子代理,串联上下文图谱)+超级基座的形式。

或许,IT部门不仅不会消失,反而要承担起更重、也更高维度的责任——转变为“AI基础设施的构建者、BP的赋能者与全面治理体系的设计者”。

IT部门的出路应该是奔向Super BI(Business intelligence), 让老板用最便宜的方式看到可辅助决策的实时数据, 彰显贡献!

智能体走向自迭代与反思层?

技术底座的演进自然是绕不开的话题。来自红熊AI等底层技术大咖的分享,为全场数字化操盘手打开了未来6个月的技术天窗。

在讨论到大模型最容易发生的“上下文污染”和记忆混乱卡点时,红熊AI的专家指出,单纯依赖拉长上下文(Context Window)是无法彻底解决模型精度下降问题的,目前的行业突破口在于Memory(记忆机制)与上下文的隔离,从而保持模型的清晰度。

更令人振奋的是关于“智能体自我迭代与反思层(Reflection Layer)”的探讨。当智能体节点达到成千上万时,系统的清洗、运行效率会急剧变缓。红熊AI目前正在推行“AI的自我创建与自进化形态”:利用反思层不断往回推,在运行完后自动进行提纯和重构,压榨并提升精度。

“也许在不久的未来,固定形态的Agent压根就不存在了。所有的模块都应该是根据临时的任务自动组装,就像我们今天这场闭门会,大家聚到一起,讨论完了、任务结束了,这个实体就自动解散。这才是下一代大模型技术对智能体形态的颠覆。”

观察:从案例崇拜到解法共创

此外,参会的所有嘉宾都分享了近期最新实践案例、这个实践的初衷、成效、优化方向。例如蒙牛的Claude尝试、德胧酒店AI替代前台及AI运营实践、龙湖千丁数科的AI人才分类等(部分案例将在近期陆续释出,诚邀关注)。

我最深的感受是:技术演进的速度几乎是以天为单位在狂飙,数字化的操盘手们每天都被无数的技术选择、架构迭代和无休止的投资回报率(ROI)考核顶到前线。在这个局里,大家都迫切地暴露最真实的卡点,共创最务实的解法。

“有没有案例”不再是核心问题,“落地方式是否可迁移”、“该如何解决”才是关键。每家企业的数据基础、业务规模、组织结构各异,单一的成功叙事参考价值有限。

这场闭门会的价值,不在于给出了标准答案,而在于把“权力让渡比写代码更难”“IT要完还是转型”“基层是数据生产者”这些真实判断,从私下抱怨变成了桌上共识。

同时,虎嗅智库正在从“观察者”转向“行业判断与共识形成平台”。

未来半年,我们持续聚焦营销服、运营自动化、制造供应链三大主题,每季度举办2-3场线下闭门会及线上圆桌,与产业一线一同拆解真实落地的分水岭。年底,我们将遴选最具落地实力的企业,沉淀为“大鲸榜”的核心依据。

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

AI Agent大规模落地后,IT部门面临哪些挑战和转型方向?

AI Agent大规模落地后,IT部门面临业务部门自行开发应用带来的“存在感”挑战。其转型方向是从传统的需求交付者,转变为AI基础设施的构建者、业务伙伴(BP)的赋能者以及全面AI治理体系的设计者,并可能向提供实时决策支持的Super BI(商业智能)方向发展。

企业如何衡量AI应用是否实现了有效落地?

企业衡量AI应用是否有效落地,关键看是否触发了企业级层面的真实改变。具体标准包括:是否催生了新的组织结构或调整了旧组织;是否实现了人员精简与组织瘦身;以及工作流、流程和SOP是否发生了根本性变化。若这三点均未发生,则单点AI工具在企业治理层面被视为无效。

当前AI Agent技术面临哪些主要卡点,未来的演进趋势是什么?

当前AI Agent技术面临“上下文污染”和记忆混乱等卡点。行业突破口在于通过Memory(记忆机制)与上下文隔离来保持模型精度。未来趋势是智能体走向自进化,通过引入“反思层”实现任务的自我提纯与重构,并可能发展为任务驱动的临时自动组装形态,而非固定实体。

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

家居零售商B&Q第三方平台业务GMV上涨29% 向家居全品类推进

家居零售品牌B&Q旗下第三方市场新增摄像头及可穿戴科技品类,GMV同比大涨29%,线上渗透率提升,集团CEO表示将严格管控成本并把握增长机遇。