广告
加载中

懂知识、能执行 滴普科技选择做更“Work”的Agent

杨丽 2026-07-24 09:09
杨丽 2026/07/24 09:09

邦小白快读

EN
全文速览

本文介绍了滴普科技在WAIC 2026推出的可落地干活的企业级智能体Agent产品,核心解决当前大模型行业只重聊天展示、不产出实际生产结果的痛点,核心干货如下:

1. 产品核心架构为Deepexi企业本体大模型+DeepWorks企业智能体平台,构建了“数据到知识再到执行”的完整闭环,区别于从办公助手切入的通用Agent,核心壁垒是知识体系基于企业现场业务逻辑,经过真实产线验证。

2. 技术核心逻辑分为两层,底层是本体大模型构建可动态迭代的活知识网络,理清企业各类数据的关联关系,解决传统静态知识体系过时、AI不会推理的问题;上层是Skills执行单元,让AI能落地完成具体业务。

3. 该方案已经在激光装备制造企业试点验证,实现产能提升60%、效率提升3倍、故障复发率降低30%,现已联合华为云开启云服务,中小企业也可低门槛使用。

本文内容对品牌商布局AI实现降本增效,有这些干货参考:

1. 当前企业AI应用已经从概念追捧转向落地产出实际结果,品牌商布局AI时要重点关注Token投入的ROI,优先选择能产出可交付业务结果的方案,避免盲目跟风概念造成成本浪费。

2. 品牌商可依托企业级Agent沉淀自身的业务知识,比如把导购经验、供应链运维知识、用户运营经验转化为动态迭代的知识网络,搭配对应Skills落地到知识管理、运营决策、供应链维护等场景,提升全链路运营效率。

3. 滴普科技的模式显示,现在中小企业也能低门槛用上成熟的企业级Agent能力,不同规模的品牌商都可以尝试落地AI,通过“人+Agent”的模式放大单个员工的生产效率,实现降本增效。

本文围绕企业级Agent产业最新发展,给To B服务类卖家带来这些干货信息和机会提示:

1. 当前企业级AI的竞争焦点已经转向能落地执行、为生产结果负责的Agent,市场需求从通用大模型转向可落地的产业AI,卖家可避开通用聊天Agent的红海赛道,转向深耕行业场景的落地型Agent,抓住新的增长机会。

2. 滴普科技的商业模式值得学习:通过8年服务沉淀2000多个可复用的行业Skills,降低服务同类客户的边际成本,实现“一个人带一群Agent干十个人的活”,小团队就能支撑千万级年收入,大幅提升人效。

3. 机会提示:当前市场既有头部企业的深度场景需求,也有中小企业的低门槛使用需求,卖家可参考滴普科技的做法,联合云厂商开放云服务,覆盖中小客户群体,拓展自身业务边界。

本文介绍的企业级Agent方案,给制造工厂推进数字化和AI转型带来这些干货启示:

1. 当前制造工厂普遍存在数据孤岛、知识沉睡的问题,大量生产维修经验存于老技术员脑中,传统静态数据治理体系维护成本高、无法解决真实业务问题,基于本体大模型构建动态知识网络是全新的可行解决路径。

2. 该方案的落地价值已经得到验证:在激光装备制造企业落地后,实现产能提升60%、效率提升3倍、故障复发率降低30%,可应用在设备故障维修、生产工艺优化等核心生产场景,直接解决工厂的实际痛点。

3. 工厂落地该方案门槛较低,不需要从零开始搭建AI能力,可基于已经沉淀好的制造行业Skills集快速适配自身需求,现在中小企业也能通过云服务低门槛接入,适合不同规模的制造工厂做数字化升级。

当前企业级AI服务行业呈现新的发展趋势,给AI技术服务商带来这些干货内容:

1. 当前客户的核心痛点已经转变,企业不再满足于只会聊天的通用大模型,核心诉求是Token消耗能换来可交付的生产结果,需要能执行、可治理、对企业结果负责的Agent能力,这是服务商当前最核心的市场机会点。

2. 传统数据治理体系已经无法适配AI时代需求,传统静态知识体系是给人看的,不是给机器用的,存在知识过时、维护成本高、无法推理解决真实问题的痛点,滴普科技的本体大模型+动态Skills方案可作为参考解决方案,构建从数据到知识再到执行的完整闭环。

3. 规模化拓展可参考该模式:沉淀行业本体数据和可复用Skills降低边际服务成本,同时联合云厂商开放云服务,覆盖中小客户群体,有效拓展市场规模。

本文透露出当前企业对AI平台的核心需求,以及AI平台布局的最新风向,核心干货如下:

1. 企业对AI平台的核心需求已经从提供大模型调用能力,转向支撑Agent落地生产业务的完整操作系统,平台需要提供从知识底座到执行入口再到行业能力集的完整产品体系,才能满足企业真实的业务需求。

2. 平台产品架构可参考滴普科技的分层设计:将日常工作入口、企业知识底座、高质量数据和行业Skills集分层,既方便员工日常使用,又能持续沉淀企业知识,还能通过复用行业Skills降低企业接入成本,适配企业真实使用流程。

3. 风向提示:企业Agent正在从办公助手进入行业深度场景,平台需要提前布局和国产算力厂商的合作,开放云服务降低中小客户使用门槛,同时要深耕行业场景,构建经真实业务验证的能力体系,打造自身壁垒。

本文透露出当前企业级Agent产业的最新发展动向,呈现了新的技术范式和商业模式,核心干货如下:

1. 产业新动向:当前企业级AI竞争已经从大模型参数比拼转向落地能力竞争,Token成本高企倒逼整个行业转向追求可量化ROI的落地型Agent,企业需求从通用AI转向能深度结合行业业务的产业AI,发展路径从个人办公助手走向组织级业务落地。

2. 新的技术范式:区别于传统大模型外挂RAG知识库的路径,滴普科技提出本体大模型构建动态知识网络+动态Skills执行单元的新范式,解决了传统静态知识体系过时、AI无法推理执行的痛点,该范式已经得到试点验证,具备行业推广价值。

3. 商业模式创新:To B AI服务领域,通过沉淀可复用的行业本体和Skills降低边际服务成本,同时通过云服务模式规模化覆盖中小企业,实现小团队支撑大收入的高效模式,为To B AI服务领域提供了新的研究样本。

返回默认

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 introduces Dipu Technology's newly launched enterprise-grade Agent product at WAIC 2026, a solution built for real-world production use that addresses a key pain point in the current large model industry: most existing solutions focus on conversational demos rather than delivering tangible production outcomes. Key details are as follows:

1. The product is built on the core architecture of Deepexi Enterprise Ontology Large Model + DeepWorks Enterprise Agent Platform, forming a complete closed loop "from data to knowledge, to execution". Unlike general-purpose Agents built starting from office assistants, its core moat lies in its knowledge system, which is grounded in on-site enterprise business logic and validated by real production lines.

2. Its core technical logic has two layers. The bottom layer, built on the ontology large model, creates a dynamically iterable living knowledge network that maps relationships between all types of enterprise data, solving the problems of outdated traditional static knowledge systems and weak AI reasoning capabilities. The upper layer consists of Skills execution units that enable AI to complete concrete business tasks on the ground.

3. The solution has been piloted and validated at a laser equipment manufacturing enterprise, delivering a 60% increase in production capacity, a 3x improvement in efficiency, and a 30% reduction in recurrent equipment failures. It is now available as a cloud service in partnership with Huawei Cloud, enabling small and medium-sized enterprises to adopt it with a low entry barrier.

This article offers the following actionable insights for brands looking to leverage AI for cost reduction and efficiency gains:

1. Enterprise AI adoption has shifted from chasing hype to delivering tangible on-the-ground outcomes. When deploying AI, brands should prioritize measuring the ROI of token investment, and select solutions that deliver deliverable business outcomes, avoiding waste from blindly following unproven AI trends.

2. Brands can use enterprise-grade Agents to codify their own proprietary business knowledge: for example, converting导购 experience, supply chain operation knowledge, and user operation expertise into a dynamically iterable knowledge network. Paired with dedicated Skills, this solution can be deployed to scenarios including knowledge management, operation decision-making, and supply chain maintenance to boost end-to-end operational efficiency.

3. Dipu Technology's model proves that mature enterprise-grade Agent capabilities are now accessible to small and medium-sized enterprises with a low entry barrier. Brands of all sizes can test AI deployment, leveraging a "human + Agent" model to amplify individual employee productivity and achieve cost reduction and efficiency growth.

This article outlines the latest developments in the enterprise-grade Agent industry, and shares the following insights and opportunity alerts for B2B service sellers:

1. Competition in the enterprise AI space is now shifting toward Agents that can execute on the ground and take accountability for production outcomes. Market demand has pivoted from general-purpose large models to industry-aligned actionable AI. Sellers can avoid the red ocean of general-purpose conversational Agents, and instead focus on industry-specific deep-scenario actionable Agents to capture new growth opportunities.

2. Dipu Technology's business model offers a useful reference: after 8 years of service delivery, the company has built a library of over 2,000 reusable industry Skills, which lowers the marginal cost of serving similar clients. This model enables "a single expert leading a team of Agents to do the work of 10 people", allowing a small team to support tens of millions in annual revenue and dramatically improve personnel efficiency.

3. Opportunity alert: The current market has both deep scenario demand from large enterprises and low-barrier demand from small and medium-sized enterprises. Sellers can follow Dipu Technology's example by partnering with cloud providers to launch cloud services, covering the SMB segment and expanding their business boundaries.

The enterprise-grade Agent solution introduced in this article offers the following key insights for manufacturing facilities advancing digital and AI transformation:

1. Most manufacturing facilities currently face problems of data silos and dormant knowledge: a large volume of production and maintenance expertise is locked in the heads of senior technicians, while traditional static data governance systems carry high maintenance costs and fail to solve real-world business problems. Building dynamic knowledge networks based on ontology large models offers a new, proven solution to this problem.

2. The on-the-ground value of this solution has already been validated: after deployment at a laser equipment manufacturer, it delivered a 60% increase in production capacity, a 3x improvement in efficiency, and a 30% reduction in recurrent failures. It can be applied to core production scenarios including equipment troubleshooting and production process optimization to directly address factories' core pain points.

3. The solution carries a low adoption barrier for factories: manufacturers do not need to build AI capabilities from scratch, and can quickly adapt to their own needs based on a pre-built library of manufacturing industry Skills. Small and medium-sized factories can now access the solution via cloud service with a low entry barrier, making it suitable for manufacturing facilities of all sizes pursuing digital upgrades.

The enterprise AI service industry is seeing new development trends, and this article shares the following key takeaways for AI technology service providers:

1. Clients' core pain points have shifted: enterprises are no longer satisfied with general-purpose large models that only handle chat, and their core demand is to get deliverable production outcomes in exchange for token consumption. They need Agent capabilities that can execute, are manageable, and take accountability for business results – this is currently the biggest market opportunity for service providers.

2. Traditional data governance systems can no longer meet the demands of the AI era. Traditional static knowledge systems are built for human use, not machine processing, and suffer from outdated information, high maintenance costs, and an inability to reason through real problems. Dipu Technology's ontology large model + dynamic Skills solution offers a proven reference that builds a complete closed loop from data to knowledge to execution.

3. Providers looking to scale can reference this model: codify industry ontology data and reusable Skills to lower marginal service costs, and partner with cloud providers to offer cloud services to reach small and medium-sized clients and effectively expand market size.

This article outlines enterprises' core demand for AI platforms and the latest trends in AI platform strategy, with key takeaways as follows:

1. Enterprises' core demand for AI platforms has shifted from providing large model inference access to a complete operating system that supports Agent deployment for production business. Platforms need to provide a complete product stack from knowledge base to execution entry to industry capability libraries to meet enterprises' real business needs.

2. Platforms can reference Dipu Technology's layered architecture design: separating daily work entry, enterprise knowledge base, high-quality data, and industry Skills into distinct layers. This design not only facilitates daily use by employees, but also enables continuous enterprise knowledge accumulation, and lowers adoption costs for enterprises by reusing industry Skills, aligning with enterprises' actual workflows.

3. Trend alert: Enterprise Agents are moving beyond office assistants to deep industry scenarios. Platforms should proactively build partnerships with domestic computing power providers, offer cloud services to lower adoption barriers for small and medium-sized clients, and deepen industry expertise to build business-validated capability systems and form core competitive moats.

This article shares the latest developments in the enterprise-grade Agent industry, presenting new technical paradigms and business models with key insights as follows:

1. New industry trends: Competition in enterprise AI has shifted from competing on large model parameter counts to competing on deployment and execution capabilities. Rising token costs have pushed the entire industry toward outcome-driven, ROI-focused Agents. Enterprise demand has pivoted from general-purpose AI to industry AI deeply integrated with business workflows, and development has moved from personal office assistants to organizational-level business deployment.

2. New technical paradigm: Unlike the traditional approach of attaching a static RAG knowledge base to a large model, Dipu Technology has introduced a new paradigm: ontology large model to build dynamic knowledge networks, paired with dynamic Skills execution units. This solves the long-standing problems of outdated static knowledge systems and poor AI reasoning and execution capabilities, and the paradigm has been validated in pilot deployments, making it suitable for industry-wide adoption.

3. Business model innovation: In the B2B AI service space, the approach of codifying reusable industry ontology and Skills to lower marginal service costs, while using a cloud service model to scale reach to small and medium-sized enterprises, enables small teams to support large revenue streams. This provides a valuable new research sample for the B2B AI service industry.

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.

企业智能体平台DeepWorks已经在产线上干活了。

WAIC 2026现场,滴普科技的展台前围满了人。很多人是来自各自的企业代表,他们来不是看大模型聊天的,而是来看一个Agent(智能体)究竟是怎样在企业里干活的。

这是一个基于Agent实现故障维修的工业场景片段:

当输入一条设备故障的线索后,系统并非简单的检索答案,而是基于企业已经构建好的知识模型,自主规划出完整的故障排查路径。先查什么数据、如何判断、按照什么逻辑推进,每一步都可追溯。如果你还想让系统自动生成一段诊断代码,或者调用模型生成一段操作视频供技术员参考,也同样可以,甚至未来还可能将维修方案直接对接给机器人执行。

这个场景,恰好揭开了2026年企业级AI竞争焦点的冰山一角。当Token越烧越贵成为业界共识时,更多的客户开始追问:这些Token消耗,能否换来可交付的生产结果?

企业真正需要的,不是只会陪你聊天作答的AI,而是能执行、可治理、且能够为企业结果负责的Agent能力。滴普科技近期升级的Deepexi企业本体大模型+DeepWorks企业智能体平台,正是针对这一需求的直接回应。

读懂企业

先读懂数据背后的“关系”

同样还是以工业场景举例。

走进任何一家有一定体量的制造工厂的IT与数据分析部门,你会看到各种各样的数据不断产生:设备传感器每秒吐出的振动频谱、三维扫描仪生成的点云数据、CAD图纸里的几何参数、ERP系统里的订单流转数据,还有企业内部日常办公积累的大量工作数据……但问题是,这些数据之间缺乏统一的语义关联和关系协作。

举个例子,振动频谱异常和轴承磨损之间有什么关系?温度曲线漂移和光学元件老化为什么有关?这些知识藏在老技术员的脑子里,散落在一条条的流程日志和故障手册里,还有不同系统里的一个个数据孤岛。

AI不是看不懂数据,而是还没看懂数据之间的关系。

“关系”为什么重要?在此之前,企业一定做过类似的事情:花了大力气去做数据治理,建立数据和知识体系,然后训练AI,让AI去理解工作流,或者让大模型外挂一个RAG知识库。走到这一步你会发现,这个数据体系永远是静态或过时的,AI依然只能检索已有的数据,依然回答不了你正在面临的真实业务问题。因为它始终是给人看的,不是给机器用的。没有AI的持续运转,知识的维护成本远大于使用收益,最终只能“沉睡”在文档库里。

滴普科技创始人赵杰辉有个很直白的理解:数据本身很重要,但原来的数据工具平台和数据治理方法论在AI时代已经不太适用了。企业需要的是能够持续迭代和进化的知识网络,并且还需要形成知识网络为核心的Skill组合方式。

这是本体大模型在做的事情。

滴普科技构建的本体大模型,是给工厂建立了一套体系化的知识体系。不是把故障手册扫描成PDF喂给AI,那样AI只会检索,不会推理。而是把维修这件事,变成一张活的知识网络:故障现象、可能原因、排查步骤、所需工具、历史案例、相关设备参数、关联工艺节点。这些节点之间的关系,是AI真正读懂工厂的关键。

区别在于,过去的知识图谱是给人看的,而现在的知识图谱是给Agent做的。Agent可以持续运转、持续迭代,知识体系也就自然能够将价值落地下去。

在某家激光装备制造企业,这套知识网络已经建立起来了。从设备、工艺、老师傅的经验,都源源不断被编译到一套口径一致的知识网络。当技术员报出一个故障线索,部署在企业智能体平台上的DeepSense工业智能体不是去文档里搜关键词,而是基于激光设备的本体知识网络,做动态的长程任务规划:排查传感器、调出历史案例、生成修复脚本,就像一个在车间里干了二十年的老师傅。

这家企业的试点数据验证了这套机制的价值:实现产能提升60%、效率提升3倍、故障复发率降低30%。

从固定工作流到动态Skills

8年攒下2000多个“手艺”

在与赵杰辉的交流中,他还反复提及了另一个关键词:Skills。

什么是Skills?如果说知识网络是AI的大脑,Skills就是AI的手,没有Skills,AI智能看懂问题但做不了事情,有了Skills,AI可以把能力落地到具体的业务单元。

这也是滴普科技第一阶段在做的事情:依靠固定工作流,人先把步骤设计好,模型按步骤执行即可。这个过程,滴普科技花了8年时间、服务了400多家头部客户的基础上,靠手工积累了2000多个覆盖制造、消费、公共事业等领域可以复用的原始Skills。

而现在这个阶段则不再依靠工作流,变成了:瞄准企业某个领域的知识体系,让模型基于知识体系做动态任务规划,而Skills则是这个规划的执行单元。

这意味着每一次服务客户,滴普科技都会积累新的行业本体和Skills。积累越多,服务下一个同类客户的边际成本就越低,质量就越高。

在Deepexi大模型上,承载的企业场景正在持续扩展:从最初的设备维修,到如今更核心的生产工艺优化,从工业领域逐步向更多行业渗透。

今年,滴普科技把这套能力进行深度技术改造,并升级了产品体系:DeepexiOS AI级企业操作系统,定位为企业Token生产力平台,将DeepWorks作为日常工作入口,Deepexi作为企业知识底座,Deepology作为高质量数据与行业Skills集。

DeepWorks是企业员工每天打开的工作界面,但它不是又一个聊天机器人,它承载着企业级Agent的一整套工程化机制:Harness负责工具调用的策略检查、隔离执行、过程观测和结果回灌;Loop支持多智能体分工、协同、复盘和持续执行;知识中心、API、MCP、插件、审批流程则能够让AI能够进入企业真实业务流程,而不是停留在生成答案层面。

Deepexi是企业大模型,负责把日常工作数据转换成本体范式的知识逻辑。先存在Foil模块里,稳定后通过后训练进入模型权重,让企业的活字典越用越厚,形成真正的企业知识底座。

Deepology是高质量企业本体数据集和2000多个Skills的集合,也是滴普科技沉淀行业Know-how的关键底座。但这些Skills不是静态的功能模块,而是一个个最小的执行单元。Deepology企业本体数据和Skills集的专家团通常具备面向某个领域工作的基本初始配置,它们既包含任务方法、业务规则、工具接口,也承载特定场景下的专业判断逻辑。基于Deepology,企业可以快速组建具备领域能力的AI专家团,而不是从零训练一个“小白员工”。例如DeepSense生产制造Skills集、DataSense运营决策Skills集、FDESense IT服务Skills集等。这些企业本体数据和Skills集共同构成了AI员工进入具体岗位前的基本初始配置。

赵杰辉说,滴普科技现在全球有20余个办事处,每个办事处只有3到5个FDE工程师,每年需要完成的年收入都在千万以上。如果传统To B服务的逻辑,三五个人肯定做不下来。但现如今,在滴普科技每个FDE的背后,都有一批Agent员工在支撑,从项目管理、解决方案,到数据治理、资料整理。

这其实就在说明,尽管FDE模式在很多人看来是耗费人力必须做重的,但滴普科技已经在通过自身实践证明:将自身Agent化,不是让Agent替代人,而是让一个人带着一群Agent,干原来十个人的活,放大这种生产效率。

往更多实用产业AI中去

如果说激光装备制造企业的案例验证的是单个场景的深度价值,那么滴普科技面临的更大挑战,是如何把这种能力从一家头部企业复制到更多行业、更多规模的企业中去。

在这方面,硅谷的Palantir用二十年时间证明,深度服务政府和大企业的数据加AI模式可以创造数千亿美元价值。滴普科技在走一条相似但更艰难的路,中国企业更碎片化的需求、更严苛的成本意识,倒逼它必须找到更高效的规模化路径。

从今年开始,DeepWorks开始逐步开启云服务能力,与华为云合作,基于国产算力构建部署能力。原来只有头部企业才能享受的AI能力,现在中小企业在办公协同、知识管理、代码开发、业务运营等场景里,也能低门槛使用。

从产业趋势看,企业Agent服务正在从研发工具、办公助手,逐步进入组织协同和行业场景。如今,市面上已经发布了一些产品,它们代表了不同入口的企业级Agent探索。能够看到,与市面上不少从办公助手、个人助理切入的Agent产品不同,DeepWorks的知识体系建立在企业现场的数据和业务逻辑之上,Skills形成于真实产线的反复验证。这种“从数据到知识再到执行”的闭环,是它区别于通用型Agent产品的核心壁垒。

赵杰辉说,滴普科技希望基于DeepexiOS AI级企业操作系统的产品组合,在企业领域形成面向办公协同、代码能力、行业知识和深度业务任务的产业化Agent服务能力。可能对于滴普科技而言,今天在这一范式上的率先投入,也会转化为其他玩家短期内难以逾越的护城河。

伴随Agent走进千行百业,滴普科技要做的,就是让Agent真正转化为生产力,让每一笔来自企业的Token投入,都换成可交付的生产结果。这恰恰是Agent实现ROI的关键路径,也是企业推进Agent落地的最后一公里。

注:文/杨丽,文章来源:钛媒体(公众号ID:taimeiti),本文为作者独立观点,不代表亿邦动力立场。

文章来源:钛媒体

广告
微信
朋友圈

FAQ回顾

企业级智能体(Agent)和普通聊天AI有什么不同?

企业级Agent可基于企业知识网络自主规划业务执行路径,支持工具调用、多智能体协同,可对接企业真实业务流程交付生产结果。而普通聊天AI仅能提供信息检索与问答,无法直接落地到业务环节创造生产价值。

滴普科技的企业级Agent产品体系包含哪些部分?

滴普科技构建了DeepexiOS AI级企业操作系统矩阵,包含作为日常工作入口的DeepWorks企业智能体平台、作为企业知识底座的Deepexi本体大模型、承载2000+行业可复用Skills的Deepology数据集三大核心模块。

企业使用智能体(Agent)能获得什么实际效果?

从滴普科技服务的激光装备制造企业试点数据来看,落地企业级Agent可实现产能提升60%、运营效率提升3倍、设备故障复发率降低30%,同时可减少无效Token消耗,提升企业AI投入的ROI。

滴普科技的Agent服务覆盖哪些行业和场景?

滴普科技的Agent服务已覆盖制造、消费、公共事业等领域,可适配设备维修、生产工艺优化、运营决策、IT服务等多类业务场景,同时已开放云服务能力,支持中小企业低门槛使用相关AI能力。

这么好看,分享一下?

朋友圈 分享

APP内打开

+1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0