广告
加载中

千峰对话:产业智能体会不会替换平台的价值链地位

亿邦动力 2026-09-10 17:20
亿邦动力 2026/09/10 17:20

邦小白快读

EN
全文速览

这篇产业智能体主题的前沿对话,透露出不少和普通人工作、投资相关的关键信息与实用认知,值得重点了解。

1.当前智能体已经走过早期概念探索阶段,界面布局形成全行业统一共识,接下来将进入全面竞争、市场价值重新分配的阶段,未来可能出现平台给智能体打工的局面,各类平台、应用软件、专业人士都可能成为智能体生态里的可调用模块,不再是直接对接用户的核心入口。

2.对个人职业发展的实用启示是,单纯靠存量的通用专业知识已经守不住竞争壁垒,不管是企业还是个人,都要往一线实际业务场景、具体操作流程扎根,形成和实际场景深度绑定的不可替代能力,才不会被通用AI替代。

3.投资相关的实用提醒是,当前算力相关产品存在明显的价格炒作,高位入场风险很高,按照技术迭代规律,未来算力成本大概率会大幅下降,不要盲目追高相关资产。

品牌商需要重点关注产业智能体带来的价值链重构趋势,提前调整经营布局,避免在新一轮价值分配中被边缘化。

1.渠道布局要适配新的入口格局,智能体未来会抢占核心用户入口,传统电商、服务平台可能逐步沦为智能体的内置技能插件,品牌不能只盯着传统平台做渠道投放,要提前适配智能体生态的连接器、技能模块规则,保住用户触达的核心通道。

2.核心竞争力的构建逻辑要调整,单纯靠行业知识、标准化运营流程建立的壁垒会被通用大模型快速击穿,品牌要把核心优势往实际业务场景、全链路运营流程、终端服务层下沉,把业务数据、场景流程和AI能力深度绑定,建立难以被复制的护城河。

3.智能工具的应用不能停留在表面,不要只用通用智能体做PPT、出报表这类办公提效工作,要把智能能力嵌入成本核算、定价决策、渠道对接的全业务流程,结合内部经营数据和外部市场行情做决策,自动对接上下游资源,形成经营闭环。

卖家要清晰识别产业智能体发展带来的风险与新机会,及时调整经营思路,抓住新的增长红利。

1.首先要留意潜在风险,未来智能体将成为核心用户入口,传统电商平台可能成为智能体的内置功能,如果卖家只依赖传统平台的流量,未来可能在价值分账体系中只能获得很少的收益,甚至被淘汰;同时单纯靠标准化运营经验建立的优势会失效,通用大模型可以快速生成常规的店铺运营方案。

2.要抓住新的合作机会,垂直产业智能体生态会释放大量合作空间,只要卖家在细分领域有实打实的服务能力,能成为智能体生态里的优质服务商、技能提供方,就能以新的方式获得稳定流量;未来A to A的自动交易模式会逐步普及,卖家要提前做好商品、服务信息的标准化,适配智能体自动询价、竞价、下单的交易规则。

3.经营决策上可借鉴智能体的逻辑,不要只跟着外部市场行情做判断,要结合自身成本、库存、资金链情况综合决策,同时理性看待算力投入,不要在价格高位盲目采购相关设备。

工厂可以从产业智能体的发展趋势中,明确生产端数字化升级的方向,挖掘新的商业增长机会。

1.生产数字化升级要避开表面化的误区,不要只满足于用通用大模型做报表、做方案这类表层提效,要把智能系统扎进生产场景、作业流程、硬件设备层,让智能系统能直接对接生产设备、传感器,结合内部生产数据、库存数据、外部订单和行情数据,自动调整生产参数、生成执行工单,实现感知、决策、执行、结算的全链路闭环,从原来人找系统的模式,转向系统找人、系统直连设备的高效模式。

2.要抓住新的产品需求机会,未来产业智能体需要打通生产端的硬件节点,能适配智能体直连调度的生产设备、产业专用机器人会有大量需求,工厂可以提前布局软硬智一体的生产设备,对接垂直产业的智能操作系统,成为产业生态里的核心硬件节点。

3.内部组织管理可参考智能时代的调整逻辑,把内部生产流程、工艺规则API化,让AI可以直接调用提效,把更多人力投向工艺研发、客户深度服务,把技术工人的一线经验沉淀成可规模化复用的组织能力。

服务商需要认清产业智能体带来的行业格局变化,找准新定位,打造能真正解决客户痛点的服务方案。

1.要把握明确的行业发展趋势,当前智能体界面布局已经形成行业共识,即将进入全面竞争阶段,未来通用智能体会抢占核心用户入口,如果软件、服务类厂商只做带独立登录入口的通用产品,很可能逐步沦为通用智能体的插件,甚至被直接替代,单纯靠输出通用行业知识的服务,会被通用大模型快速击穿壁垒。

2.要抓准客户的核心痛点,当前很多企业应用通用大模型,只能实现做PPT、整理资料这类表层办公提效,没法真正帮企业跑通业务、提升收益、降低成本,企业也不敢把核心业务流程、生产设备直接对接通用大模型,怕出现决策失误、生产事故。

3.解决方案要往垂直深度走,采用“基础大模型+行业小模型+规则引擎”的架构,沉淀行业知识底座、实操技能库,对接企业原有业务系统,帮客户实现从数据调取、决策分析到流程执行的端到端服务,同时要理性布局算力,不要在价格高位盲目囤积相关资源。

平台商需要警惕产业智能体对现有价值链地位的冲击,及时调整发展战略,巩固自身的核心价值位势。

1.要提前规避发展中的方向性风险,当前智能体界面布局已经定型,即将进入全行业价值重分配阶段,如果平台只做表层的流量撮合交易,很可能沦为智能体生态里的一个普通功能插件,在后续价值分账中只能拿到小部分利润,甚至被替代;同时要意识到,单纯靠行业信息差、通用知识建立的平台优势,会被通用大模型快速击穿。

2.要明确平台深耕的核心方向,必须往垂直产业的纵深扎根,不能只停留在流量交易层,要向下扎进产业的实际场景、业务流程、设备操作系统层,做支撑产业智能体运行的底层操作系统,打通数据、流程、设备、交易的全链路,形成感知、决策、执行、结算的闭环,建立通用大模型无法替代的产业厚度。

3.生态建设和运营上,可以参考成熟实践,搭建API化的开放底座,把生态伙伴的能力封装成可调用的技能库,对接企业内部系统、外部交易资源、终端硬件设备,甚至结合专用机器人形成软硬一体的生态,同时算力投入要保持理性,避免在价格高位盲目囤货造成损失。

本次对话披露了产业智能体发展阶段的多个新动向、新商业模式与待研究的新问题,具备较高的产业研究价值。

1.产业发展新动向方面,当前智能体已经走过概念探索期,界面布局形成全行业共识,功能框架基本定型,即将进入全面竞争、价值重分配阶段,未来可能出现平台给智能体打工的价值链重构现象:智能体抢占用户核心入口,平台、软件、专业人才分别成为智能体的可调用模块。大模型竞争正从上半场的通用能力比拼,转向下半场的产业纵深比拼,竞争高地从云端转向产业现场,纯知识壁垒已无法阻挡通用大模型渗透,只有扎进企业场景、流程、设备层的主体才能建立护城河。

2.新商业模式方面,垂直领域可采用“基础大模型+行业小模型+规则引擎”的架构,打造“大数据+大模型+专家服务”的产业智能体,实现从决策到执行、结算的全链路闭环,甚至直连硬件形成A to A的生产、交易协同;企业内部也在向API化方向重构,经验从个人禀赋转化为可规模化的组织能力。

3.待研究的新问题方面,当前算力价格存在明显炒作泡沫,国产GPU与海外产品仍有性能差距,算力投资风险、智能体直连生产设备的安全责任界定等问题,都需要配套政策与规则引导。

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 cutting-edge discussion on industrial AI agents reveals critical, practical insights relevant to ordinary people’s work and investment decisions, and is well worth close attention.

1. AI agents have moved past the early conceptual exploration stage, with an industry-wide consensus now formed around their core interface design. The sector is set to enter a phase of full-scale competition and value redistribution. In the future, platforms may end up working for AI agents: a wide range of platforms, software applications, and professional workers could become callable modules within the agent ecosystem, rather than serving as core user-facing entry points.

2. A key takeaway for individual career development is that existing generic professional knowledge alone can no longer sustain a competitive moat. Both companies and individuals need to anchor themselves deeply in frontline business scenarios and concrete operational workflows, building irreplaceable capabilities tightly bound to real use cases to avoid being displaced by general-purpose AI.

3. A practical investment warning: there is obvious speculative price inflation in current computing power-related products, creating high risks for investors entering at peak valuations. Following standard technology iteration cycles, computing power costs are highly likely to drop substantially in the future, so investors should avoid blindly chasing high-priced related assets.

Brands must pay close attention to the value chain restructuring driven by industrial AI agents, and adjust their operational layouts in advance to avoid being marginalized in the coming round of value redistribution.

1. Channel strategies must adapt to the new entry point landscape. AI agents will capture core user access in the future, while traditional e-commerce and service platforms may gradually become built-in skill plugins for agents. Brands cannot focus solely on channel marketing on traditional platforms; they need to adapt early to the connector and skill module rules of the agent ecosystem to retain core channels for reaching users.

2. The logic of building core competitiveness must shift. Moats built solely on industry knowledge and standardized operational processes will be quickly eroded by general-purpose large models. Brands should sink their core strengths into real business scenarios, full-chain operational workflows, and end-user service layers, deeply binding their business data and scenario processes to AI capabilities to build hard-to-replicate competitive barriers.

3. AI tool adoption cannot remain superficial. Brands should not limit general-purpose agent use to office efficiency tasks such as creating presentations or generating reports. Instead, they should embed AI capabilities across full business workflows including cost accounting, pricing decisions, and channel coordination, combine internal operational data with external market intelligence to drive decisions, and automatically connect upstream and downstream resources to form closed-loop operations.

Sellers need to clearly identify both risks and new opportunities brought by the development of industrial AI agents, adjust their operational strategies in a timely manner, and capture new growth dividends.

1. First, watch for potential risks: AI agents will become the core user entry point in the future, and traditional e-commerce platforms may evolve into built-in functions of agents. If sellers rely solely on traffic from traditional platforms, they may end up receiving only a tiny share of value in the new revenue split system, or even be eliminated entirely. Meanwhile, advantages built on standardized operational experience will become ineffective, as general-purpose large models can quickly generate routine store operation plans.

2. Seize new cooperation opportunities: the vertical industrial agent ecosystem will open up extensive collaboration space. As long as sellers have solid service capabilities in niche segments and can become high-quality service providers or skill contributors in the agent ecosystem, they can gain stable traffic through new channels. As agent-to-agent (A2A) automated transaction models become increasingly common, sellers should standardize product and service information in advance to adapt to agents’ automated inquiry, bidding, and ordering rules.

3. For operational decision-making, draw on the logic of agents: do not make judgments solely based on external market trends, but integrate internal factors including costs, inventory, and cash flow for holistic decisions. In addition, take a rational view of computing power investment, and avoid blind procurement of related equipment at peak prices.

Factories can identify clear directions for production-side digital upgrading from industrial AI agent development trends, and tap into new business growth opportunities.

1. Avoid superficial pitfalls in production digital transformation: do not settle for surface-level efficiency gains from using general-purpose large models to generate reports or draft plans. Instead, embed intelligent systems deep into production scenarios, operational workflows, and hardware equipment layers, enabling smart systems to directly connect to production equipment and sensors. By combining internal production and inventory data with external order and market data, these systems can automatically adjust production parameters, generate work orders, and achieve a full closed loop of perception, decision-making, execution, and settlement, shifting from the old model of people seeking out systems to an efficient model where systems route tasks to people and connect directly to equipment.

2. Capture new product demand opportunities: future industrial AI agents will need to connect to hardware nodes on the production side, creating massive demand for production equipment and dedicated industrial robots that support direct agent scheduling and control. Factories can plan ahead for integrated software-hardware-intelligence production equipment, connect to vertical industrial intelligent operating systems, and become core hardware nodes in the industrial ecosystem.

3. For internal organization and management, reference the adjustment logic of the intelligent era: turn internal production processes and craft rules into APIs that AI can directly call to improve efficiency, reallocate more human resources to process R&D and in-depth customer service, and codify frontline experienced workers’ know-how into scalable, reusable organizational capabilities.

Service providers need to recognize the industry structure changes brought by industrial AI agents, identify new positioning, and build service solutions that can truly solve customer pain points.

1. Grasp the clear industry development trend: a cross-industry consensus on agent interface design has already formed, and the sector is about to enter a phase of full-scale competition. In the future, general-purpose agents will capture core user entry points; if software and service vendors only build generic products with independent login portals, they are likely to gradually become plugins for general-purpose agents, or even be directly replaced. Services that rely solely on delivering generic industry knowledge will see their barriers quickly broken down by general-purpose large models.

2. Pinpoint core customer pain points: many enterprises currently using general-purpose large models only achieve surface-level office efficiency gains such as making presentations or organizing materials, with no real support for running core business, increasing revenue, or reducing costs. Enterprises are also reluctant to connect core business processes and production equipment directly to general-purpose large models, for fear of decision errors or production accidents.

3. Build solutions with deep vertical expertise, adopting an architecture of "foundation large model + industry-specific small model + rule engine". Accumulate an industry knowledge base and practical skill library, connect to enterprises’ existing business systems, and deliver end-to-end services covering data retrieval, decision analysis, and process execution for clients. At the same time, plan computing power deployment rationally, and avoid blind hoarding of related resources at peak prices.

Platform operators need to guard against the impact of industrial AI agents on their current value chain position, adjust development strategies in a timely manner, and consolidate their core value advantage.

1. Preempt directional risks in development: agent interface design is now finalized, and the industry is about to enter a phase of full value redistribution. If platforms only focus on superficial traffic-matching transactions, they are likely to become ordinary functional plugins in the agent ecosystem, receiving only a small share of profits in subsequent value splits, or even being replaced. It is also critical to recognize that platform advantages built solely on industry information asymmetry and generic knowledge will be quickly eroded by general-purpose large models.

2. Define the core direction for platform deep cultivation: platforms must root themselves deeply in vertical industries, rather than remaining at the traffic transaction layer. They should reach down into real industrial scenarios, business workflows, and equipment operating systems, build the underlying operating system that supports industrial agent operation, connect the full chain of data, processes, equipment, and transactions, form a closed loop of perception, decision-making, execution, and settlement, and build industrial depth that general-purpose large models cannot replicate.

3. For ecosystem building and operations, reference proven practices: build an API-based open foundation, encapsulate ecosystem partners’ capabilities into a callable skill library, connect to enterprise internal systems, external transaction resources, and end hardware devices, and even integrate dedicated robots to form an integrated software-hardware ecosystem. At the same time, maintain rationality in computing power investment to avoid losses from blind hoarding at peak prices.

This discussion reveals multiple new trends, emerging business models, and unresolved research questions in the development of industrial AI agents, carrying high value for industrial research.

1. Regarding new industrial development trends: AI agents have moved past the conceptual exploration stage, with an industry-wide consensus formed around interface design and a basically finalized functional framework. The sector is about to enter a phase of full-scale competition and value redistribution, which may lead to value chain restructuring where platforms effectively work for agents: agents will capture core user entry points, while platforms, software, and professional talent each become callable modules within the agent ecosystem. Large model competition is shifting from the first half of general-purpose capability rivalry to the second half of deep industrial penetration, with the competitive battleground moving from the cloud to on-the-ground industrial sites. Pure knowledge-based barriers can no longer block the penetration of general-purpose large models; only entities rooted in enterprise scenarios, processes, and equipment layers can build sustainable moats.

2. Regarding new business models: vertical sectors can adopt an architecture of "foundation large model + industry-specific small model + rule engine" to build industrial agents integrating "big data + large models + expert services", achieving a full closed loop from decision-making to execution and settlement, and even connecting directly to hardware to enable agent-to-agent (A2A) production and transaction collaboration. Internal enterprise structures are also being restructured around API-based access, transforming individual experiential know-how into scalable organizational capabilities.

3. Regarding new questions requiring further research: there is currently obvious speculative泡沫 in computing power prices, performance gaps remain between domestic GPUs and overseas products, and issues including computing power investment risks and the definition of safety liability for agents directly connected to production equipment all require supporting policy and regulatory guidance.

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.

【亿邦原创】近日,在北京2026世界机器人大会期间,亿邦动力董事长、亿邦智库院长郑敏,与农信互联集团董事长薛素文,深度交流了产业智能体与产业互联网领域的前沿话题。

郑敏:去年智能体还是个盲人摸象的概念,把腿脚耳朵当做各家的“大象”。年初也只是听见龙虾热,没真看见龙虾满地跑。今天,国内外各家大厂拿出的“大象”倒是都长得差不多了。智能体界面布局初步形成了行业共识,最左侧一列功能栏,公网问答、私域助理;技能、专家、连接器生态。右侧一个对话框,中间是用户交互区。很像当年的门户网站首页布局一个样,后来的零售电商网站,首页布局也是一个样。规律在智能体身上重演,界面布局定型之后必然是全面竞争和快速迭代,市场会重新分配价值。如果说现在有点儿"商户给平台打工"的感觉,接下来是不是会有"平台给智能体打工"的局面?

薛素文:我完全认同。界面布局的趋同,本质上是功能与价值导向趋于一致的体现。您刚才说的这个"平台给智能体打工"的判断很有意思,能具体展开吗?

郑敏:知识平权+生态网络,当前智能体已经很有全链全景全要素协同的“智能生态网络”感觉。智能体抢占用户入口,各种相关平台只能变成了其中的某个Skill,各方面牛人只能成为AI分身里的某个等待召唤的“专家”,各种应用软件都是连接器的某个外挂。

薛素文:这个趋势我看到了,最近有人提出软件行业的“断头论”,即许多软件厂商,逐步去掉自家的登入界面,转而成为通用智能体的某种插件或API,与这种说法差不多,但这样的趋势并不意味着这些公司或专家就不行了,就是换一种生存与发展的方式而已,关键还是你这个“专家”是不是某个领域的“真专家”。

郑敏:您认为智能体最终会不会替代平台,成为价值链的顶级分配者?毕竟,平台经济的出现也就二三十年,一定程度上抢了大型集团或连锁企业的价值链分配权。

薛素文:平台会不会被替代,我觉得关键在于垂直领域能不能扎得足够深。AI的竞争也是一样:大模型的上半场拼的是模型能力,下半场拼的是产业纵深,高地不在云端,在产业现场。我曾经以为行业知识壁垒很坚固——比如我们多年打磨的产业级系统,我让通用大模型说"你把整个行业的业务流程梳理出来",它很容易就生成出来了,跟我这个干了多年的"专家"做的有差别,但不大。只要是我以前做的东西,通用大模型也知道,这是很值得让人思考的现象。这说明单纯的知识守不住壁垒,真正的护城河,是模型要扎进企业的场景、流程与设备等操作系统,所以,产业互联网平台必须继续向下扎,扎到操作系统这一层,否则确实会有危机。

郑敏:行业大数据公司可能也很危险。通用大模型公司会找做行业数据的公司合作——"把数据给我,我流量大,可以推出行业数据收费服务",用户付费调用数据消耗TOKEN,我给你分账。但在分账的同时,很可能就形成了垂直分工关系,通用大模型公司拿大头,行业大数据公司还得竞争着分小账,否则连小钱都分不到。我非常同意您做深的观点。超级科技大厂会不停研发出更锋利的刀片,从面上在一层一层地向下片共性利润。产业互联网公司只有加速度往下扎,做出产业厚度。否则,刀片片到现有价值面的时候,产业互联网平台就被片没了。

薛素文:这个比喻很形象。对此,我们可以达成一个共识:纯粹靠知识壁垒,很难对抗大模型的吞噬。业务是根,数据是土,AI是翼——无论多么专业的知识,都要与企业的场景、流程深度绑定,才可能长成护城河。具体到农信数智的实践,我们的爱思AI大脑的支撑底座—爱思大模型,已经正式通过国家网信部门备案,它采用"基础大模型+行业小模型+规则引擎"的架构:懂通用语言,又精农牧专业,还守得住行业规则的底线。它底下垫着三块底座——行业知识底座,让AI懂行业;专家SKILL,把多年服务产业沉淀的技能封装成AI会干活的技能库;IAP技术底座,是运行了11年的数智农业操作系统,让AI扎得住根。通用模型的智能体可以深入企业的办公领域,但很难扎进企业的业务领域,在我们看来,你可以用通用智能体做一个很炫酷的PPT或报表,但如果不能替企业跑业务、多打粮、降成本,终究还是停留在“提效率”这个领域。

我们的智能体多了些不一样的东西:从企业原有系统里主动调取数据,结合市场出现什么数据行情动向,爱思大模型和智能体反向触发企业流程化执行。这是一个先内观再外观,避免完全被市场带偏节奏,更有决策主导能力的新机制。一句话说,我们做的是"大数据+大模型+专家服务"的爱思大脑,不是从外面贴上去的通用插件。

我给您举一个"该不该卖猪"的完整例子——这不是一个单点功能,而是产业操作系统在一个产业里端到端跑通的样本。智能体的新决策流程是:第一,调取最近几个月生猪的成本数据(来自企业原有系统);第二,检查资金链状况;第三,分析行业价格趋势。如果爱思大模型决策分析认为——价格会降、现金流紧张、库存不多——那就必须卖。第四,怎么卖,智能体会指导你直接将要买的猪上架“国家生猪市场”,并将相关信息定向同步给能够收猪的“猪经济”或屠宰企业,如果实在没人要了,公司的前端服务人员还会利用公司掌握的行业资源,为其牵线搭桥。但这只是开始。以前决策就是"哦,卖吧",现在不是。爱思大脑智能体会自动下卖猪工单到ERP流程,自动上架商城;如果有人来询价,智能体自动回答;如果有人竞价,智能体判断该卖给谁。将来有可能买方也是个Agent,A to A,一整套。如果卖掉了,钱进来了,回到传统软件做收款付款,最终出报表。

在这之前,智能体还会"绕过你的思想"。这批猪要不要吃料?它还会驱动饲喂器去调整配方,自动喂养。以前饲喂系统由场长掌控,场长通过饲喂器的配置系统,为猪只设置饲喂方案,其本质上还是取决于场长个人的“经验”,现在大模型和智能体绕过场长的思想了,直接跟机器对话,跟饲喂系统对话,饲喂系统又跟传感器对话。比如,饲喂系统会根据最近一段时间猪只采食情况、猪场里的环境参数、料仓或存库里的库存余料、机器人回传料槽中剩料的情况、甚至是当前的行情、公司账上的现金情况,自动调整饲喂曲线。配置最贴近“现实”的饲喂策略,而不是用通识理论勾画出来最“科学”的饲喂建议,这背后是两重重构:流程上,从"人找系统"变成"系统找人";生产关系上,从"人畜对话"变成"人机协同"——感知、决策、执行、结算一整条链,第一次由同一个产业操作系统闭环完成。这条闭环在生猪产业里跑通了,就可以复制到任何有数据厚度的产业。

郑敏:是的,很难想象养猪场的饲喂器或者工厂数控机床敢直接按照通用大模型生成的指令干活儿,谁家饲喂器、数控机床也不敢直连豆包、千问、元宝、GPT们。产业互联网向下扎,势必要打到硬件设备层。薛总,那AI给你们企业内部组织管理带来了什么变化?

薛素文:我们公司把所有代码都“API”化了——以前是写给人看的,现在写给AI看。过去一个人有十天的工作,现在可能一个人一个小时就干完了。代码研发人员肯定可以大幅缩减,公司每年花出去的代码费用,也要顺应趋势做结构性调整,把更多资源投向行业知识工程的研发与农业AI深度服务上。上个月,公司正式上线了“天工系统”,并在此基础上成立了一个新部门--FDA服务中心(FDA即:Fast Design、Development Agility、Auto Delivery),将公司最懂行业、最懂用户、最懂AI的那一拨人组成一个“独立的组织”,在天工平台上为用户提供“从需求到产出”的闭环服务,在服务理念上,我们与市场上广泛推崇的“FDE”不一样,我们更加强调基于客户实际,推行“快速设计、敏捷开发与自主交付”的AI服务理念,本质上,这是能力的重构:经验正从稀缺的个人禀赋,变成可复制、可规模化的组织能力。

我这次在北京世界机器人大会听了很久,之前也接触了不少机器人企业,具身智能软硬件进展都很快。实体产业场景里的专用机器人,是我正在重点看的方向。我们的布局是"爱思AI大脑×农芯AI机器人"——大模型与机器人结合,从"大脑"到"手脚",构建覆盖农牧全产业链的AI生态全景。当智能体不只能调度流程、还能直接调度设备时,"产业操作系统"才真正长出了手脚,形成软硬智一体的闭环优势。总之,通用大模型只会提升我们的能力,是我们当下扎进行业的利器,它不是威胁而是机遇。用一句话说,通用大模型,让我们扎得进农场,上得了厅堂。

郑敏:以产业互联网为地基,盖好产业智能体的时候,盈利模式会不会变成Token付费主导,产业互联网公司有没有更底层的算力服务收费机会?

薛素文:Token付费是显而易见的,能不能主导要看各家能力。算力投资方面,我个人的建议是要偏冷静地看,因为现在价格炒太高了。英伟达的DGX B300在美国卖五十万美金,大概人民币三百多万,在国内炒到一千四五百万,而且还买不到。国产GPU服务器我也弄了几个,当前确实还不能完全跟上,从性价和渲染能力差距还比较大。即便如此,高点入场也是挺有风险的。摩尔定律之下,谁也不敢保证算力成本不会大幅度下降,就像当年手机话费1块钱一分钟,现在大概是一两毛吧。

郑敏:那是不是在将来,每个垂直产业智能体也会长得差不多?

薛素文:界面布局的逻辑类似,但结合产业深度,会有各自的打法和独特壁垒。我们对自己的定位,就是做产业智能体背后的产业操作系统——向下扎进数据、流程和设备,向上长出智能体。产业互联网,也确实还可以有个更年轻的名字,"产业智能体"是很准确的。也很赞同你们在第八届亿邦产业互联网年会的基础上,叠加了"首届产业智能体大会"。

编者按:薛素文,现任北京农信互联科技集团公司董事长、农信数智股份有限公司董事长、总裁。他主导了大北农20年信息化及农业+互联网业务,并创办农信互联,目前专注于农信数智公司农业AI服务平台建设,曾获评2010年度中国十大优秀CIO等奖项、行业优秀创业家、担任北京农业互联网协会理事长、中国畜牧业协会智能畜牧分会会长等社会职务。在薛素文先生的领导下,农信数智荣获2025年度千峰奖 "产业AI 30强",该奖项是亿邦动力、亿邦智库持续跟踪研究产业互联网领域,持续推出的高专业度奖项。

本文首发于 亿邦动力 官方网站

文章来源:亿邦智库

广告
微信
朋友圈

FAQ回顾

产业智能体会取代产业互联网平台的价值链主导地位吗?

产业智能体界面布局已形成行业共识,未来将抢占用户交互入口,传统互联网平台、应用软件、行业专家可能成为其内置技能、插件或可调用模块;但产业互联网平台只要向下深耕企业场景、流程、设备层,构建产业纵深壁垒就不会被替代。

产业智能体时代垂直产业平台的核心护城河是什么?

单纯的公开行业知识无法形成长期竞争壁垒,垂直产业平台真正的护城河是将AI与企业实际业务场景、流程、硬件设备深度绑定,打造覆盖感知、决策、执行、结算全链路闭环的产业操作系统,形成难以被通用大模型渗透的产业厚度。

农业产业智能体有哪些可落地的实际业务价值?

以农信数智爱思AI大脑为例,农业产业智能体可实现农牧场景端到端业务闭环:自动分析养殖成本、资金链、行情数据给出售猪决策,联动完成交易、结算全流程,还可对接饲喂设备动态调整饲喂方案,直接帮养殖企业降本增效。

产业互联网企业布局AI需要注意哪些风险?

产业互联网企业布局AI需规避两类风险:一是避免单纯依赖公开行业知识构建壁垒,这类知识很容易被通用大模型快速掌握;二是算力投资需保持理性,当前高端GPU价格被大幅炒高,摩尔定律下算力成本存在快速下降可能,高点入场风险较高。

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