广告
加载中

腾讯云副总裁吴运声拆解企业AI账单:最贵的往往不是Token

胡镤心 2026-07-22 18:07
胡镤心 2026/07/22 18:07

邦小白快读

EN
全文速览

这篇文章整理了2026世界人工智能大会上,腾讯云副总裁吴运声关于企业级AI落地的核心干货内容,核心要点如下:

1. 当前AI行业竞争已经从比拼大模型参数,转向实际落地应用,Agent取代大模型成为行业最热方向,企业AI投入成本不止大家关注的Token,系统改造、员工协同磨合、数据治理、安全合规才是占比更高、更影响AI价值创造的投入。

2. 2026年企业级Agent同比增长至少翻倍,行业目前还处于开局阶段,增长空间大,普通用户已经被AI广泛教育,市场教育成本大幅下降,技术也能支撑更多之前做不了的业务落地。

3. 本次大会腾讯云发布了ADP 4.0智能体开发平台海外版,面向出海企业做了全方面的国际化适配,出海已经成为AI行业的必选项。

本文对品牌布局AI、把握AI落地趋势、优化AI投入有较高参考价值,核心干货如下:

1. 当前AI已经进入落地阶段,Agent是行业主流方向,C端用户已经被AI充分教育,市场教育成本低,品牌布局AI的时机已经成熟,信息化基础好的品牌落地速度会更快。

2. 品牌布局AI要算全成本账,不能只看Token成本,系统改造成本、数据治理成本、安全可控要求才是决定AI能否创造价值的核心,布局时不用推倒原有老系统,做适配性迭代即可,控制投入。

3. 目前Agent已经在客服、营销等品牌常用场景成熟落地,品牌可以采用Agent+Workflow双向结合的模式,开放模糊需求用Agent,确定性流程用Workflow,既能满足业务需求,又能有效控制AI投入成本,同时出海已经成为必然趋势,品牌可依托适配海外生态的平台快速落地AI。

本文给布局AI的卖家指明了行业趋势、增长机会与潜在风险,核心干货如下:

1. 当前企业级Agent行业处于早期开局阶段,同比增长至少翻倍,赛道增长空间大,C端用户已经被AI充分教育,市场教育成本低,技术也能支撑各类业务落地,属于难得的增长红利期,卖家可积极切入。

2. 卖家布局AI可采用Agent搭配不同模型、拆分任务的模式降本,开放模糊需求用Agent处理,确定性流程用Workflow执行,不需要反复消耗Token,能有效提升AI投入的ROI,同时对接原有系统不用完全推倒重建,做适配性改造即可,降低改造成本。

3. 当前出海已经成为卖家AI布局的必选项,东南亚市场是重点布局方向,卖家可以依托ADP 4.0海外版这类适配海外生态的平台,快速搭建AI应用,不需要额外操心基础设施,大幅降低出海AI落地的门槛。

本文给工厂推进数字化AI转型带来了不少启示和商业机会,核心干货如下:

1. 当前AI行业已经进入落地应用阶段,企业级Agent技术已经成熟,能够支撑工厂的数字化转型需求,而且市场教育成本低,整个社会对AI的接受度很高,工厂转型AI的门槛在持续降低,有前瞻性的传统工厂也可以走在转型前列。

2. 工厂布局AI转型不需要完全推倒原有的老信息系统,只需要做适配性的迭代改造就能满足Agent对接需求,不需要完全重建,大幅降低了转型的投入成本,也减少了转型对现有生产的影响。

3. Agent技术已经在工业巡检这类工厂常用场景落地,过去需要针对不同场景单独训练模型,还要做数据后处理结构化,现在只需要用智能体搭建一个管线就能解决问题,能有效降本提效,工厂可以优先在这类场景尝试落地AI,拿到转型成效。

本文清晰梳理了当前企业级智能体行业的发展趋势,以及企业客户的核心痛点,对AI服务商调整业务方向有较高参考价值,核心干货如下:

1. 行业发展趋势方面,当前AI竞争已经从比拼模型参数转向落地应用,Agent成为行业核心热点,企业级Agent今年同比增长至少翻倍,行业还处于早期开局阶段,增长空间极大,同时出海已经从可选项变成必选项,海外市场存在大量未被满足的需求。

2. 客户核心痛点方面,企业最关注的不是Token降价,而是现有老系统对接难题,多数老系统没有成熟的调用接口,原本是为人操作设计,无法适配Agent需求,同时企业对AI的可信、可控、可审计有极高的安全要求。

3. 解决方案方向可参考腾讯云的思路:打造Agent与Workflow双向互调功能,帮客户拆分任务控制整体成本,针对出海客户做好海外渠道、主流大模型、常用SaaS的生态适配,满足企业出海的全方面需求。

本文分享了腾讯云智能体开发平台的最新动态,总结了企业对智能体平台的核心需求,给同类平台发展指明了方向,核心干货如下:

1. 企业客户对智能体平台的核心需求可总结为三点:一是能够兼容对接企业现有老系统,支持适配性改造,不需要企业推倒原有系统,承担高额改造成本;二是能够帮助企业控制AI整体成本,同时满足企业对AI可信、可控、可审计的安全要求;三是出海客户需要平台完成全链路国际化适配,接入海外主流渠道、大模型和SaaS生态。

2. 平台产品迭代可参考ADP 4.0的升级方向:新增Agent与Workflow双向互调功能,适配开放场景和确定性流程的不同需求,同时推出专门的海外版本,围绕触达、交互、生态、连接完成国际化适配,重点布局东南亚市场,抓住中国企业出海和全球AI落地的红利。

3. 当前智能体行业还处于发展早期,平台不需要过早严格切分业务边界,可先迭代发展,后续再逐步形成稳定的行业格局。

本文记录了当前企业级AI产业的最新动向,以及头部平台玩家的核心判断,对产业研究有较高的参考价值,核心干货如下:

1. 产业最新动向方面,AI行业竞争已经从比拼模型参数转向实际落地应用,Agent取代大模型成为行业最热方向,2026年企业级Agent同比增长至少翻倍,行业目前仍然处于开局阶段,远未到终局,Token价格持续下降,但企业AI成本结构发生变化,系统改造、数据治理、安全合规等成本占比远高于Token成本。

2. 商业模式方面,按任务计费因为任务难以精确定义,目前头部平台并未采用,主流模式仍然是围绕Token计价,通过Agent+Workflow结合的方式帮助客户控制整体成本,提升AI投入ROI,目前ADP平台已经落地30多个行业,多个场景都已经跑出成熟案例。

3. 产业新趋势方面,出海已经成为中国智能体产业的重要战略,中国在智能体落地过程中积累的工程化能力、场景化经验、性价比优势,正在向外溢出,变成全球企业可享受的基础设施红利,东南亚是当前国内智能体平台出海的重点布局区域。

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 summarizes key insights from Tencent Cloud Vice President Wu Yunsheng’s speech on enterprise AI implementation at the 2026 World Artificial Intelligence Conference, with core takeaways as follows:

1. The AI industry’s competition has shifted from competing on large model parameter sizes to real-world implementation. Agents have replaced large models as the hottest direction in the space. Beyond the widely discussed token costs, enterprises’ AI investment is dominated by other items that have a far greater impact on AI value creation: system upgrades, employee collaboration alignment, data governance, and security & compliance.

2. Enterprise-grade Agent adoption will grow by at least 100% year-over-year in 2026. The industry is still in its early opening phase with enormous room for growth. End consumers have already been broadly educated on AI, cutting market education costs dramatically, while existing technology can now support implementation of more use cases that were previously unfeasible.

3. At the conference, Tencent Cloud launched the overseas version of its ADP 4.0 Agent Development Platform, with full internationalization adaptations for cross-border enterprises. Going global has become a mandatory strategic move for the entire AI industry.

This article offers high-value reference for brands looking to布局 AI, capitalize on AI implementation trends, and optimize AI investment, with key takeaways as follows:

1. AI has now entered the phase of large-scale implementation, and Agents have become the mainstream industry direction. End consumers are already fully familiar with AI, bringing down market education costs significantly. The timing is right for brands to start their AI布局, and brands with solid digital foundations will see faster implementation.

2. Brands need to account for full lifecycle costs when布局 AI, rather than only focusing on token costs. System transformation costs, data governance costs, and requirements for secure, controllable AI are the core factors that determine whether AI can deliver business value. Brands do not need to discard their existing legacy systems; adaptive iterative upgrades are sufficient to control investment.

3. Agents have already achieved mature implementation in common brand use cases such as customer service and marketing. Brands can adopt a combined Agent + Workflow model: use Agents for open-ended, ambiguous demands, and Workflow for deterministic processes. This meets business requirements while effectively controlling overall AI investment costs. Meanwhile, going global is now an inevitable trend, and brands can leverage platforms adapted to overseas ecosystems to implement AI quickly.

This article clarifies industry trends, growth opportunities and potential risks for sellers布局 AI, with core insights as follows:

1. The enterprise-grade Agent industry is still in its early opening phase, with projected year-over-year growth of at least 100% and massive headroom for growth. End consumers are already fully educated on AI, cutting market education costs, and current technology can support implementation across a wide range of business use cases. This is a rare period of growth dividend, and sellers can actively enter the space.

2. To cut costs when布局 AI, sellers can adopt a model that combines Agents with different models and splits task types: Agents handle open-ended, ambiguous demands, while Workflow executes deterministic processes. This avoids unnecessary repeated token consumption and effectively improves the ROI of AI investment. Additionally, sellers do not need to rebuild their existing systems from scratch; adaptive transformation is sufficient to lower integration costs.

3. Going global has become a mandatory step for sellers’ AI布局, with Southeast Asia emerging as a key target market. Sellers can leverage platforms adapted to overseas ecosystems such as the overseas version of ADP 4.0 to build AI applications quickly, without worrying about underlying infrastructure, which greatly lowers the barrier to AI implementation for cross-border expansion.

This article offers valuable insights and business opportunities for factories advancing digital AI transformation, with core takeaways as follows:

1. The AI industry has now entered the phase of real-world implementation, and enterprise-grade Agent technology is mature enough to support factories’ digital transformation needs. With low market education costs and high societal acceptance of AI, the barrier to AI transformation for factories continues to decrease, allowing forward-thinking traditional factories to take the lead in transformation.

2. Factories do not need to discard their existing legacy information systems for AI transformation. Adaptive iterative upgrades are sufficient to enable Agent integration, eliminating the need for full system rebuilding. This significantly cuts transformation investment costs and reduces disruption to ongoing production during the transition.

3. Agent technology has already been implemented in common factory use cases such as industrial inspection. In the past, factories had to train separate models for different scenarios and complete post-processing and structuring of data; now, a single pipeline built with Agents can solve these problems, effectively cutting costs and improving efficiency. Factories can prioritize testing AI implementation in these types of scenarios to deliver early transformation results.

This article clearly outlines current development trends of the enterprise-grade Agent industry and the core pain points of enterprise clients, offering high-value reference for AI service providers adjusting their business strategies, with core insights as follows:

1. On industry trends: AI competition has shifted from competing on model parameters to real-world implementation, with Agents emerging as the core industry hotspot. Enterprise-grade Agent will grow at least 100% year-over-year this year, and the industry remains in an early opening phase with enormous growth potential. Meanwhile, going global has shifted from an optional strategic choice to a mandatory one, with large unmet demand in overseas markets.

2. On core client pain points: Enterprises’ top concern is not token price cuts, but the challenge of integrating Agents with existing legacy systems. Most legacy systems lack mature call APIs and were originally designed for human operation, making them incompatible with Agent requirements. Enterprises also have extremely high security requirements for trusted, controllable and auditable AI.

3. On solution directions, Tencent Cloud’s approach offers a useful reference: build bidirectional interoperability between Agents and Workflows to help clients split tasks and control overall costs; and build full ecological adaptations for cross-border clients, including support for overseas channels, mainstream large models and popular SaaS tools, to meet all the requirements of enterprises expanding overseas.

This article shares the latest updates on Tencent Cloud’s Agent development platform, summarizes enterprises’ core requirements for Agent platforms, and outlines a clear direction for peer platforms, with core insights as follows:

1. Enterprise clients’ core requirements for Agent platforms can be summarized in three points: first, the platform must be compatible with enterprises’ existing legacy systems and support adaptive transformation, so enterprises do not need to discard existing systems and bear high reconstruction costs; second, the platform must help enterprises control overall AI costs while meeting requirements for trusted, controllable and auditable AI; third, cross-border clients require full end-to-end international adaptation, with access to overseas mainstream channels, large models and SaaS ecosystems.

2. For platform product iteration, the upgrade direction of ADP 4.0 offers a reference: add bidirectional interoperability between Agents and Workflows to adapt to the different requirements of open scenarios and deterministic processes; launch a dedicated overseas version with full internationalization adaptation covering reach, interaction, ecosystem and connectivity, and prioritize布局 in the Southeast Asian market to capture the dividend from Chinese enterprises’ global expansion and global AI implementation.

3. The Agent industry is still in an early development stage, so platforms do not need to strictly define business boundaries prematurely. They can prioritize product iteration first, and gradually form a stable industry structure over time.

This article documents the latest developments in the enterprise AI industry and core judgments from a leading platform player, offering high reference value for industrial research, with core insights as follows:

1. On the latest industry developments: AI industry competition has shifted from competing on model parameters to real-world implementation. Agents have replaced large models as the hottest industry direction, with enterprise-grade Agent growth of at least 100% year-over-year in 2026. The industry is still in its opening phase, far from reaching a mature end state. While token prices continue to fall, the cost structure of enterprise AI has shifted: system transformation, data governance, and security and compliance costs account for a far larger share of total spending than token costs.

2. On business models: Due to the difficulty of precisely defining individual tasks, leading platforms have not adopted pay-per-task pricing. The dominant pricing model remains token-based, with providers helping clients control overall costs and improve AI investment ROI through a combined Agent + Workflow approach. To date, the ADP platform has been implemented in more than 30 industries, with mature use cases across multiple scenarios.

3. On new industry trends: Going global has become a core strategy for China’s Agent industry. The engineering capabilities, scenario-based experience and cost advantages Chinese players have accumulated in Agent implementation are now spilling outward, becoming infrastructure benefits available to enterprises globally. Southeast Asia is currently the key focus market for Chinese Agent platforms expanding overseas.

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】2026年世界人工智能大会现场,Agent取代大模型成为出现频率最高的词汇。AI竞争已经从比拼模型参数,转向了实际落地应用。

可以看出,Token降价了,但企业AI的总账单反而更复杂了——系统改造的投入、员工协同的磨合、数据治理的成本、安全可控的底线……这每一笔,都比Token单价更复杂,也更能决定一个智能体到底是在炫技,还是在创造价值。

大会期间,腾讯云正式发布了智能体开发平台ADP 4.0海外版。腾讯云副总裁吴运声和我们聊了聊企业级智能体落地这一年——进展到了哪一步,卡在哪里,以及下一步往哪走。

1、增长“至少翻倍”

被问及今年企业级Agent的增长情况时,吴运声给出一个数字:同比至少翻倍。

技术本身在往前走,过去检索、工作流能力不够强,一些任务即使完成系统对接也难以稳定执行;而今年Agent Loop能力增强后,以前做不了的事情在技术上已经可以支撑。其次,市场教育成本在快速下降——客户在日常工作和生活中已经被AI“教育”过了,不需要再解释“什么是智能体”。

但增长的另一面是挑战。吴运声特别提到一个被很多人忽略的难点:怎么对接企业已有的老系统。

“很多数据或者流程都沉淀在各个老系统里,有些信息化做得不是很好的企业,老系统也没有成熟的MCP接口或者API可以调用。”他说。这就需要对系统做适当改造,或者提供接口让Agent能够调用,才能嵌入真实业务流程。

更麻烦的是,老系统过去是为“人”设计的——界面给人看,按钮由人点,结果由人判断。但未来,系统设计理念也要考虑为Agent服务。不过,吴运声强调,完全推倒重来不是主流选择,“至少现在还没有看到太多企业敢下这个决心。更多还是做适配性的迭代、改进和更新。”

至于企业级Agent应用的核心要求,吴运声认为是两点:可信、可控。“对企业来讲,我得知道到底是谁做了什么,后面所有东西都要可审计、是安全的,这非常重要。”

至于智能体行业的终局,吴运声说得很坦然:“现在还只是开局阶段,远远没有到终局。”智能体从两三年前开始不断出现,能力不断演进,各种各样产品都会冒出来。“早期不用非常严格地切分‘这个一定做什么、那个一定做什么’,就这样发展着,可能慢慢会走向某种平衡或者终局。”

2、Token成本不是唯一账本

今年企业开始认真算AI的账了,Token成本、整体ROI都是绕不开的话题。

吴运声认为,Token成本确实在快速下降,但企业算账的方式也在进化。越来越多企业会搭配不同模型完成不同事情;也会把环节拆分——模糊的事情用Agent处理,确定性流程用Workflow来做,不需要反复消耗Token。

ADP 4.0的一个关键升级正是Agent与Workflow双向互调。这个设计的逻辑很务实。企业有些场景是开放的、意图模糊的,适合用Agent处理;有些流程是严谨的、确定性的,适合用Workflow执行。两者结合,既满足业务需求,又控制成本。

至于外界讨论的“按任务计费”模式,吴运声的回应很直接:“我们一般不会按照任务来收费,任务太难精确定义了。”

至于哪些行业跑得快?吴运声的判断是:长期来讲都会被改造,短期来看信息化做得好的行业更快。

泛互联网行业本来就出身于数字化,进入Agent很快;金融行业信息化领先,生意模式健康,也跑在前面。但他也补充,不排除一些传统行业里有前瞻性的公司,照样能走在前列。

至于跑出成效的场景,除了常见的客服和营销,吴运声提到了巡检——过去货架巡检要单独训练模型,针对不同场景训练不同模型,再把数据做后处理、结构化出来;现在用智能体搭一个管线就能解决。传媒领域也用得很多,电视台用ADP制作生成管线、电视节目,或者检索需要的媒体资源。

目前ADP平台已落地30多个行业,覆盖智能客服、知识管理、媒体生产等业务场景。

3、出海:从“可选项”变成“必选项”

ADP 4.0海外版是这次发布的重点,这不是临时起意——腾讯云此前已在海外有不少落地,某大型医疗集团利用ADP构建了覆盖智能导诊、门诊预约、急诊风险识别等全流程的AI医疗助手,预计可独立处理80%以上的常规咨询和预约。

海外版围绕“触达、交互、生态、连接”四大能力做了全面国际化适配。Claw应用支持LINE、Telegram等主流海外渠道;智能工作台支持自定义时区与多区域调度;模型广场接入多家国际主流大模型;平台深度接入Google Workspace、Confluence、Jira等海外SaaS应用。

对于海外市场的选择,吴运声说会重点看东南亚等区域,综合考虑市场人口、基建情况、政治环境和对华关系。“出海是整个中国企业非常重要的战略。”

ADP 4.0海外版接入了多家国际主流大模型,深度整合Google Workspace和Jira等海外SaaS生态,让企业不必为基础设施操心,上手就能搭。从这个角度看,出海不是“产品卖出去”那么简单。它是一种溢出——中国在智能体落地过程中积累的工程化能力、场景化经验、性价比优势,正在变成全球企业都能享受到的基础设施红利。

文章来源:亿邦动力

广告
微信
朋友圈

FAQ回顾

企业部署AI的成本除了Token还有哪些?

企业部署AI的总成本除Token费用外,还包含系统改造投入、员工协同磨合成本、数据治理成本、安全合规建设投入等,这些成本的复杂度远高于Token单价,也直接决定智能体应用的实际价值。

腾讯云ADP 4.0海外版有什么功能特点?

腾讯云ADP 4.0海外版围绕触达、交互、生态、连接四大能力做国际化适配,支持LINE、Telegram等海外渠道,接入多家国际主流大模型,深度整合Google Workspace、Jira等海外SaaS应用,可降低企业智能体落地门槛。

企业级智能体哪些行业和场景落地更快?

短期来看信息化基础好的行业落地速度更快,泛互联网、金融行业走在前列,传统行业中具备前瞻性的企业也有领先落地案例。目前已覆盖30多个行业,常见落地场景包括智能客服、知识管理、媒体生产、巡检等。

企业级智能体应用的核心要求是什么?

企业级智能体应用的核心要求是可信、可控,所有操作需要可审计、安全,企业可明确追溯操作主体与执行内容,避免不可控的业务风险,保障业务运行稳定性。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +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