广告
加载中

DAA、超级个体与混合编队:李彦宏的AI时代进化论

张帅 2026-05-14 09:55
张帅 2026/05/14 09:55

邦小白快读

EN
全文速览

本文核心是李彦宏在百度Create2026大会上提出的AI时代全新认知体系,核心干货和可参考信息如下:

1. 核心判断:AI行业的核心度量衡将从Token转为日活智能体数(DAA),Token只衡量AI投入成本,DAA衡量AI实际完成的有效任务,是AI时代生产力经济的新规则,李彦宏预测未来全球DAA有望突破100亿。

2. AI时代进化论包含三层核心内容:智能体将从被动响应进化为主动自我升级,普通人将借助AI成长为超级个体,企业会进化为人与智能体的混合编队。

3. 普通个体也能抓住AI机会,当前低门槛AI开发工具已经能让零编程基础的人把生活创意变成落地应用,比如温州二年级小学生就用秒哒做出了全校可用的校园拼伞小程序,普通人可借助工具降低开发门槛,成长为AI时代的超级个体。

本文针对AI时代品牌发展,梳理出了符合产业新趋势的干货启示,核心内容如下:

1. 产业和消费趋势已经发生变化:AI行业从过去的流量逻辑转向效率变现逻辑,品牌竞争不再是抢用户注意力,而是比拼AI落地带来的实际效率提升,只有能完成实际任务的AI应用才能创造商业价值。

2. 产品研发层面可抓住新机会:当前自然语言开发工具已经大幅降低应用创作门槛,品牌可以快速将用户的个性化需求转化为落地的小工具、小应用,快速测试市场反应,小需求也能快速落地,更贴合消费者需要。

3. 组织转型有明确方向:品牌要将组织转型为人与智能体的混合编队,用智能体承担标准化任务,人类负责创意和决策,同时要调整管理规则:多授权少管控、减少层级加快决策、提高人才密度而非人海战术,更好适配AI时代的竞争节奏。

本文为AI时代的卖家梳理了新的增长机会、商业逻辑和组织优化方向,核心干货如下:

1. 新的赛道机会:AI的主流应用形态是智能体,行业不需要扎堆卷大模型参数,应该聚焦落地应用,未来全球DAA有望超过100亿,对应海量的智能体落地需求,聚焦能实际解决问题、完成任务的智能应用就是新的增长机会。

2. 商业逻辑已经变化:互联网时代靠流量变现,AI时代靠效率变现,头部企业的案例已经验证这个趋势:Anthropic依靠企业级场景的扎实价值交付,年度经常性收入达到300亿美元反超OpenAI,说明交付价值比流量规模更重要。

3. 卖家自身组织优化方向:要转型为人与智能体的混合编队,调整管理规则:多授权少管控,避免压制创造力;减少汇报层级,加快决策迭代速度;靠更高人才密度和AI能力出结果,不用人海战术;多给AI派任务,发挥人类把控意图的独特价值,整体提升运营效率。

本文为工厂推进AI数字化转型、抓住新商业机会提供了明确方向,核心干货如下:

1. 产品生产和设计迎来新机遇:低门槛AI开发工具普及后,工厂不需要依赖外部专业开发团队,就能把一线生产、设计环节的个性化需求快速变成定制AI应用,解决自身实际问题,就像零编程基础的小学生都能做出可用小程序,工厂也能快速落地适配自身场景的工具。

2. 新的商业机会:智能体已经成为AI应用的主流形态,未来全球DAA有望突破100亿,工厂可以围绕生产端、消费端的实际任务需求,开发适配自身产业场景的智能体,切入AI配套赛道,拓展新的增长空间。

3. 数字化转型可参考成熟经验:工厂转型要走人与智能体混合编队的路线,用智能体承担标准化生产任务,人类负责创新和决策,同时调整组织架构适配AI转型,已有青岛港使用百度AI智能管控系统实现10.21%效率提升的案例,可参考其价值导向的转型思路。

本文梳理了智能体时代AI服务行业的发展趋势、客户痛点和解决方案方向,核心干货如下:

1. 行业发展新趋势:AI行业已经从拼大模型参数、比Token消耗的阶段,转向聚焦落地交付价值的阶段,DAA将成为AI行业新的核心度量衡,智能体是AI应用的主流形态,未来全球会有超过100亿日活智能体,智能体相关服务市场空间巨大。

2. 当前客户的核心痛点:过去行业扎堆做大模型,并没有解决AI落地最后一公里的问题,客户需要的是能实际完成任务、实现价值闭环交付的智能体,而非单纯的大模型能力,对能支撑智能体全链路运行的基础设施和应用开发工具需求迫切。

3. 解决方案可参考的方向:服务商可以围绕智能体的需求搭建全栈能力,参考百度芯云模体全栈架构的思路,从底层算力集群、AI云基础设施、大模型能力到上层智能体应用,打造完整能力,同时推出低门槛开发工具,帮助客户快速落地自有智能体应用,匹配市场需求。

本文点明了智能体时代平台的发展方向、需求变化和风险规避思路,核心干货如下:

1. 市场对AI平台的核心需求已经转变:传统平台是为普通应用和大模型开发服务,智能体时代平台需要为大规模智能体提供全栈基础设施支撑,客户需要的是能支撑智能体完成闭环任务的全套能力,而非单纯的算力供给。

2. 平台布局可参考成熟路径:百度的芯云模体全栈架构已经验证了方向,底层打造万片级协同的算力集群,满足智能体高并发实时运行的需求;中层升级面向智能体的AI云,提供记忆存储、任务编排、安全隔离等配套能力;上层开放低门槛开发工具,让普通用户也能开发应用,扩大平台开发者生态。

3. 运营和风险规避方向:平台要引导生态偏离无意义的模型参数内卷,鼓励开发者聚焦实际落地价值,打造符合AI效率变现逻辑的生态,同时要提前搭建支撑亿级智能体并发的底层基础设施,避免未来算力不足制约生态发展的风险。

本文梳理了AI行业最新的产业动向和系统性新认知,对AI产业研究有较高的参考价值,核心干货如下:

1. 产业发展新动向:当前AI行业已经进入智能体原生阶段,行业底层逻辑从互联网的流量变现、注意力经济,转向AI的效率变现、生产力经济,行业度量衡从传统的DAU、Token转向DAA,也就是日活智能体完成的有效任务数,Anthropic反超OpenAI就是这个趋势的标志性事件,验证了新逻辑的正确性。

2. 新的系统性产业理论:李彦宏提出了完整的AI时代进化论,形成了连贯的认知体系,包含三层进化方向:智能体从被动响应到主动自我进化,人类个体从普通个体到超级个体,企业组织从人与人分工到人与智能体混合编队,是对AI产业演进方向的系统性判断。

3. 商业模式新变化:AI产业不再追求打造超级入口级的超级App,转向去中心化的数百万级垂直有用应用,头部企业已经开始布局智能体原生的全栈基础设施,百度芯云模体全栈架构就是典型代表,低门槛开发工具让全民开发成为可能,这些都是值得研究的产业新动向。

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 Robin Li's new cognitive framework for the AI era, unveiled at Baidu Create 2026. Key actionable insights are as follows:

1. Core industry thesis: The key metric for the AI sector will shift from tokens to Daily Active Agents (DAA). Tokens only measure AI input costs, while DAA quantifies the volume of effective tasks completed by AI, establishing a new rule for productivity in the AI era. Li projects global DAA could surpass 10 billion in the future.

2. Three layers of AI-era evolution: Agents will evolve from passive responders to active self-improving systems; ordinary people can grow into "super individuals" empowered by AI; and organizations will evolve into hybrid teams of humans and AI agents.

3. Accessible opportunities for individuals: Low-threshold AI development tools now allow people with no programming experience to turn creative ideas into working applications. For example, a second-grade student in Wenzhou built a campus umbrella-sharing app for his entire school using Baidu's Miaoda platform. Ordinary people can leverage these tools to lower development barriers and become super individuals in the AI era.

This article outlines key actionable insights for brands navigating AI-era industry trends, as follows:

1. Shifts in industry and consumer trends: The AI industry has moved from a流量-focused logic to an efficiency-driven monetization model. Brand competition is no longer a battle for user attention; it now hinges on actual efficiency gains from AI implementation. Only AI applications that deliver completed real-world tasks can generate tangible business value.

2. New R&D opportunities: Natural language-powered development tools have drastically lowered the barrier to building applications. Brands can quickly turn customers' personalized needs into working small tools and applications to test market response, enabling even niche demands to be launched rapidly to better match consumer needs.

3. Clear direction for organizational transformation: Brands should restructure into hybrid teams of humans and AI agents, where agents handle standardized tasks and humans focus on creativity and decision-making. To adapt to AI-era competition, brands should revise management rules to delegate more authority, reduce bureaucracy to speed up decision-making, and prioritize talent density over headcount.

This article breaks down new growth opportunities, updated business logic and organizational optimization directions for sellers in the AI era. Key insights are as follows:

1. New market opportunities: Agents have become the dominant form of AI applications, and the industry does not need to crowd into an arms race for larger model parameters. The focus should instead shift to real-world deployments. Global DAA is projected to exceed 10 billion, corresponding to massive demand for deployed agents. Focusing on smart applications that actually solve problems and complete tasks is the key to new growth.

2. A fundamental shift in business logic: The internet era relied on traffic monetization, but the AI era runs on efficiency monetization. Industry data already validates this trend: Anthropic has surpassed OpenAI in annual recurring revenue, reaching $3 billion, by delivering solid value in enterprise use cases, proving that value delivery matters more than traffic scale.

3. Directions for internal organizational optimization: Sellers should transform into hybrid teams of humans and AI agents, and adjust management practices: delegate more authority and reduce micromanagement to avoid stifling creativity; cut reporting layers to speed up decision iteration; deliver results through higher talent density combined with AI capabilities, instead of relying on brute-force headcount; and assign more routine tasks to AI to leverage humans' unique strength in aligning on core intentions, ultimately boosting overall operational efficiency.

This article lays out a clear direction for factories to advance AI-driven digital transformation and capture new business opportunities. Key insights are as follows:

1. New opportunities for product design and manufacturing: With the widespread adoption of low-threshold AI development tools, factories can turn customized needs from frontline production and design workflows into tailored AI applications in-house, without relying on external professional development teams to solve their unique problems. Just as a student with no programming experience can build a working app, factories can rapidly deploy tools tailored to their own operational scenarios.

2. Untapped new business opportunities: Agents have become the mainstream form of AI applications, and global DAA is projected to surpass 10 billion. Factories can build agents tailored to their industrial scenarios to meet real task demands on both production and consumer ends, enter the AI supporting track, and open up new space for growth.

3. Proven best practices for digital transformation: Factories should pursue a transformation path of hybrid teams combining humans and AI agents, where agents take charge of standardized production tasks while humans focus on innovation and decision-making, accompanied by organizational restructuring to adapt to AI transformation. Qingdao Port has already achieved a 10.21% efficiency gain using Baidu's AI-powered intelligent control system, and factories can reference its value-oriented transformation approach.

This article summarizes development trends, core client pain points and solution directions for the AI service industry in the agent era. Key insights are as follows:

1. New industry trends: The AI industry has moved beyond the phase of competing on large model parameters and token consumption, and now focuses on delivering deployed real-world value. DAA will become the new core metric for the industry, with agents as the dominant form of AI applications. Global daily active agents are expected to exceed 10 billion, creating enormous market opportunity for agent-related services.

2. Core current client pain points: The industry's past rush to build large models has failed to solve the "last mile" problem of AI deployment. Clients need agents that can actually complete tasks and deliver end-to-end value, rather than just access to raw large model capabilities. There is strong unmet demand for infrastructure and development tools that support full-lifecycle agent operation.

3. Reference directions for solution development: Service providers can build full-stack capabilities tailored to agent needs, following the example of Baidu's "Core-Cloud-Model-Agent" full-stack architecture. This means building a complete capability stack from underlying computing clusters and AI cloud infrastructure to large model capabilities and upper-layer agent applications, alongside launching low-threshold development tools to help clients rapidly deploy their own custom agent applications to match market demand.

This article lays out development directions, shifting demand and risk mitigation strategies for platforms in the agent era. Key insights are as follows:

1. Shifts in core market demand for AI platforms: Traditional platforms serve general application and large model development, but in the agent era, platforms need to provide full-stack infrastructure support for large-scale agent deployment. Clients require end-to-end capabilities that enable agents to complete closed-loop tasks, rather than just raw computing power supply.

2. A proven path for platform development: Baidu's "Core-Cloud-Model-Agent" full-stack architecture has already validated this direction: it builds a 10,000-GPU collaborative computing cluster at the bottom layer to meet the high-concurrency real-time operation requirements of agents; upgrades an agent-optimized AI cloud in the middle layer to provide supporting capabilities including memory storage, task orchestration and security isolation; and opens low-threshold development tools at the upper layer to enable ordinary users to build applications and expand the platform's developer ecosystem.

3. Directions for operations and risk mitigation: Platforms should guide their ecosystems away from meaningless competition over model parameters, and encourage developers to focus on real-world deployment value to build an ecosystem aligned with AI's efficiency monetization logic. Platforms should also pre-build underlying infrastructure to support hundreds of millions of concurrent agents to avoid the risk of future computing power shortages constraining ecosystem growth.

This article summarizes the latest industry developments and a systematic new cognitive framework for the AI industry, offering high reference value for AI industry research. Key insights are as follows:

1. New industry development trends: The AI industry has now entered the native agent era. The underlying logic of the industry has shifted from the internet-era model of traffic monetization and attention economy to AI's efficiency monetization and productivity economy. The core industry metric has shifted from traditional DAU and tokens to DAA, the volume of effective tasks completed by daily active agents. Anthropic's recent surpassing of OpenAI is a landmark event for this trend, validating the correctness of this new logic.

2. A new systematic industry theory: Robin Li has put forward a complete theory of AI-era evolution, forming a coherent cognitive framework with three layers of evolutionary direction: agents evolve from passive response to active self-improvement; individual humans evolve from ordinary people to super individuals; and organizations evolve from human-only division of labor to hybrid teams of humans and agents, representing a systematic judgment on the direction of AI industry evolution.

3. New changes in business models: The AI industry is no longer focused on building super entry-level super apps, and is instead shifting to a decentralized landscape of millions of vertical useful applications. Leading players have already begun to lay out native full-stack infrastructure for agents, with Baidu's "Core-Cloud-Model-Agent" full-stack architecture as a typical example. Low-threshold development tools have also made mass citizen development possible. All of these are new industry developments worthy of further 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.

只有那些敢于打破惯性、持续重塑自身的企业,才有机会真正穿越周期,建立新的竞争优势。

AI行业从不缺共识,缺的是敢于提出“非共识”的人。真正改变产业走向的,也往往不是追逐热点的跟随者,而是那些在质疑声中率先看见下一阶段的人。

在Create 2026大会上,百度创始人李彦宏再次抛出新的判断:日活智能体数(DAA)将成为AI时代的度量衡。

这并非他第一次逆着行业风向发言。从“不要卷模型,要卷应用”,到“智能体是AI应用的最主流形态”,李彦宏的许多观点,一开始并不被多数人认同,却又在行业演进中不断被验证。

当巨头们开始涌向同一条赛道,AI竞争也越来越像一场没有终局的“无限游戏”——没有固定规则,没有最终赢家,只有不断扩展的边界与持续迭代的方向。

如果李彦宏这一次的“非共识”依然成立,那么AI行业的下一关,或许已经开始了。

AI行业喧嚣背后

一条更长期的逻辑主线

AI行业以天为单位迭代,每天都有新的技术名词诞生,每周都有新的赛道被定义。但如果我们把时间轴拉长就会发现,李彦宏的AI认知之间有一条清晰的逻辑主线,每一个新观点都是上一个的延伸。

2024年,是李彦宏密集输出“非共识”的一年。那一年,整个行业陷入了一种集体性的癫狂:巨头们疯狂堆叠算力参数,创业者们争相训练更庞大的基座模型,所有人都相信“模型即一切”。

在这样的背景下,李彦宏抛出了一个反直觉的判断:不要卷模型,要卷应用。这个观点在当时显得不合时宜,当大多数人在为“百模大战”欢呼时,他却在提醒行业关注落地的最后一公里。

紧接着,他又提出了一个非共识,——智能体(Agent)是AI应用的最主流形态,即将迎来爆发点。彼时,市场的主流叙事还在寻找下一个ChatGPT式的“杀手级App”,人们习惯性认为,AI的终局,将是一个无所不包的超级入口。

李彦宏还明确表示:“百度不是要打造超级应用,而是要打造数百万级‘超级有用’的应用。”在“超级App”思维根深蒂固的互联网下半场,这种去中心化的应用观,几乎是对传统流量逻辑的一次彻底背离。

这些判断,在2024年的语境下,是孤独的。然而,站在2026年的节点回看,行业现实正在逐一印证这些预判的准确性。

李彦宏这几年的非共识也在进化。从“人人都是开发者”,进化成了“人人都是超级个体”;从“智能体是最看好的方向”,进化成了“DAA是AI时代的度量衡”;从呼吁“内化AI能力”,到本次Create大会上,系统性地提出“AI时代的进化论”。

这条线索的连续性,正是李彦宏的判断与行业其他观点的最大区别,它不是每年换一个新概念,而是每年在一个方向上再往前推一步,直到形成一套完整的认知体系。

Agent的价值锚定点

不应该是Token

在AI行业的账本里,Token是最性感的数字。它直接代表着算力消耗,代表着模型调用的频次,也代表着模型厂商和云厂商的收入。

“Token不一定代表终局,它只代表成本并不代表收益。”在李彦宏看来,Token衡量的是投入,而非产出。显然,如果一家公司只盯着Token看,那它看到的只是成本中心,而不是价值中心。

李彦宏预测,未来全球DAA可能会超过100亿。显然,这个数字背后,不是100亿个聊天的人,而是100亿个正在被执行的、有价值的任务。

为什么DAA比Token更重要?因为它回答了一个关键问题,AI的价值到底怎么算?互联网时代的逻辑是流量变现,只要有人看,就有广告价值,AI时代的逻辑是效率变现,只有把事做成,才有商业价值。

最有力的佐证来自全球AI市场的格局变迁。2026年3月,Anthropic的年度经常性收入(ARR)达到300亿美元,正式反超OpenAI。这是一个具有标志性意义的时刻,OpenAI依然拥有更高的日活用户数(DAU),但Anthropic凭借其在企业级场景中更扎实的“交付价值”,赢得了资本市场的更高溢价。

这恰恰与李彦宏最新提出的DAA逻辑不谋而合。DAA不看有多少人跟AI“聊”过天,只看有多少智能体真正帮人把事“办”成了,以前我们数人头、看时长,那是互联网时代注意力经济的老套路,现在我们要看效率、看闭环,这是AI时代生产力经济的新规则。

但DAA不仅仅是一个财务指标,它更是一场关于“进化”的系统性变革。李彦宏提出了“AI时代进化论”,其含义包括三层,一是智能体的自我进化,从被动响应到从环境中不断汲取营养来提升自己,并主动执行;二是人类个体的自我进化,从普通个体到超级个体,并学会跟AI共存;三是企业组织的自我进化,从人与人的分工协作,到人与智能体的混合编队,成为超级组织。

首先是智能体的进化,过去的AI是被动响应的,你问它答。未来的智能体,必须具备从环境中汲取营养、自我升级的能力。

这次百度发布的通用智能体DuMate,以及近期出圈的Hermes agent,都展示了这种趋势。它们不再局限于单点任务,能够整合多种能力,自主规划路径,甚至在执行过程中发现错误并自我修正。

其次是个体的进化。本届百度create大会,在秒哒APP及企业版发布环节,一名8岁的小朋友,分享了他用秒哒创作应用的经历。

小孩子名叫扑满,来自温州,是一个二年级的小学生。他讲起了自己和同学发明的“哒哒打伞”——下雨天没带伞的人“发单”,有伞的同学“接单”,一起拼伞出校门。因为“别的班听不见”,他索性用秒哒做出了一个全校都能用的小程序。大屏幕上,一个由自然语言生成的应用界面随之展开。

这让人意识到,真正被改变的,也许不只是开发效率,而是软件的创造门槛:当一个二年级孩子,都能把生活里的小灵感变成应用,“超级个体”开始第一次拥有了真正可感知的模样。

最后是组织的进化,传统企业的组织架构,是基于人与人之间的分工建立的,在AI时代,组织将演变为“人与智能体的混合编队”,智能体承担标准化的、高并发的任务,人类负责创造性的、决策性的工作。

李彦宏表示:“自我进化是一套面向AI时代的系统性变革。只有那些敢于打破惯性、持续重塑自身的企业,才有机会真正穿越周期,建立新的竞争优势。”

企业组织具体该如何进行自我进化?在本届大会上,李彦宏也提出了四点思考,一是更多授权,更少管控。“在智能体时代,过度企业管理往往不是降低风险,而是在压制创造力。”

二是更快对齐,更少层级。“减少汇报层级,信息直达、即时决策对AI时代的组织来说,至关重要。因为,迭代速度才是企业竞争的护城河。”

三是更高人才密度,更少人海战术。“AI时代不是靠堆人提高成功率,而是靠更优秀的人、更强的AI能力、更高密度的人才配置,做出更漂亮的结果。”

四是更多任务,更少分工。“今天AI几乎是全能的,它唯一不能的是知道你的意图,它不知道你想完成什么任务。所以我们要通过提示词多给AI派任务,提示词是你的思考过程,是人的独特价值。”

“芯云模体”,百度的通关底气

李彦宏在Create2026上的一系列判断,如果只看作是对未来的预测,未免太浅。当主动调用API的是智能体,自主选择模型的是智能体,自动查询数据的也是智能体时,整个底层基础设施,必须为这个全新的“数字物种”重新搭建。

某种程度上,李彦宏之所以敢于将非共识提前公布,也是因为百度自身具备足够厚的护城河。百度的“芯—云—模—体”全栈架构,这是智能体原生时代的系统性底座。

首先是芯片,很多人对算力的理解还停留在单颗芯片的性能上。但在智能体时代,算力不再是单点突破,而是大规模协同。

百度昆仑芯P800已经完成了万片级交付,天池超节点256卡版本正式点亮,将于今年6月正式上线,百度拥有了支撑亿级智能体并发运行的硬件底座。智能体的运行是高度并发的,它们需要实时感知、快速决策、即时执行,昆仑芯集群提供了更好的AI基础设施。

其次是AI云,百度智能云已经升级为面向大规模智能体的AI全栈云,传统的云计算是为应用服务的,现在的AI云是为智能体服务的,它不仅要提供算力,还要提供智能体所需的记忆存储、任务编排、安全隔离等基础设施。

然后是模型,文心大模型持续迭代,最新的文心5.1多次登顶全球榜单,智能体能不能理解复杂的指令,能不能在多步任务中保持逻辑连贯,能不能在遇到错误时自我修正,这些都取决于底层的模型能力。

最后,是智能体本身,百度智能云旗下的通用智能体DuMate,能够自主完成跨应用、跨文件的复杂任务,实现从“理解”到“执行”的闭环交付,可在持续协作中学习你的工作习惯与任务偏好。

类似地,秒哒企业版支持多人协作、大型复杂项目分工、全业务流程闭环,真正实现了“生产级的产品交付平台”;伐谋2.0在青岛港自动化码头的实际部署中,码头智能管控系统A-TOS实现了10.21%的效率提升。

在智能体时代,单一环节的优势已经不足以构建护城河,“芯云模体”就是百度的通关底气,也是它在AI这场无限游戏中,能够穿越周期、持续进化的根本原因。

在这个由智能体重构的AI未来世界里,百度已然拿出来系统的AI时代进化论,并提前建造好了地基。现在,它要开始盖摩天大厦。

注:文/张帅,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

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