AI News

At Baidu World Conference 2026, Robin Li first proposed the "Three Principles of Organizational Evolution," saying the next act of AI is the intelligent organization

Sina TechUpdated Nov 12 22:03

AI summary

At Baidu World Conference 2026, Robin Li first proposed the "Three Principles of Organizational Evolution," shifting the "next act" of AI from the technology itself to the intelligent organization. The number of daily active agents within Baidu increased from an average of 5,000 per day to 37,000, and the effective tasks completed by agents each week rose from 84,000 to over 1 million; 4,300 Workspaces have been built and more than 20,000 organizational Skills have been accumulated. The conference also released the ERNIE 5.5 Flash preview version (280 billion total parameters, one-million-token context) and the large-model trading platform "Ciyuansuo."

Body · AI translation

Related topics top start

Topic: Baidu World Conference 2026 Related topics top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

In 2016, at the Baidu World Conference, Robin Li announced to the audience: the next act of the internet is artificial intelligence. That year, AlphaGo had just defeated Lee Sedol, and AI was still a frontier concept that needed evangelizing.

Ten years later, on November 12, 2026, still at the Baidu World Conference, Robin Li once again took the stage, and the theme still revolved around "intelligence"—but this time, he moved the camera away from the technology itself and focused on four words: intelligent organization.

Over the past decade, the rapid development of artificial intelligence has become a consensus. "Whether to use AI" is no longer the question; the question has become "how does the organization catch AI." In his speech, Robin Li pointed out a common confusion among many corporate leaders: "Why is Token spending increasing more and more, but revenue and profit are not rising along with it?" His answer was an analogy from the industrial age—these companies have only replaced the electric motor, but have not yet redesigned the factory.

How to build the new factory? Robin Li first proposed the "three principles of organizational evolution"

In Robin Li's view, the past decade has been a decade in which AI formed an independent technology stack, and Baidu has grown layer by layer around the four-layer architecture of "chip, cloud, model, agent"—chips, cloud computing, large models, and agents. If the upstream of AI technology is energy, then the downstream is organizational capability: an AI-native organizational system different from the conventional bureaucratic hierarchy. To this end, he proposed the "three principles of organizational evolution" for the first time.

First, break down job boundaries. Jobs are the unit of the industrial age; tasks are the unit of the intelligent age. The organization of the future is a task network in which people and Agents collaborate. He believes that jobs will not disappear, but will converge into two categories: Builders, who are responsible for creating new things, and Sellers, who are responsible for communicating and connecting with people and building trust.

Second, precipitate tacit knowledge. If experience exists only in employees' heads, it is merely individual capability; only when it is made explicit and precipitated onto the company platform does it constitute organizational intelligence. "This kind of organizational capability is something others cannot copy, poach, or buy. That is what a moat is."

Third, design a closed-loop process. The formula Robin Li gave is "measurable + Looping = continuous evolution." People and AI form a Loop around tasks, and the company becomes a highly automated operating organization composed of countless Loops, a system that continuously learns and evolves.

The three principles are not just empty talk. Robin Li presented Baidu's internal data as evidence: over the past year, the number of daily active agents grew from an average of 5,000 per day to 37,000, the number of effective tasks completed by agents grew from 84,000 per week to more than 1 million per week, and internally 4,300 Workspaces have been built and more than 20,000 organizational Skills have been precipitated.

Around this narrative, the conference simultaneously released multiple products: the large model trading platform "Ciyuansuo," the Wenxin large model 5.5 Flash preview version (280 billion total parameters, million-Token context), and upgrades to four major agents: Baidu Dazi, Kuku, Baidu Yijing, and Famou. Robin Li also judged that today's embodied intelligence is roughly at the development level of autonomous driving in 2016, while Apollo Go has covered 28 cities worldwide, with autonomous driving mileage exceeding 350 million kilometers.

Shen Dou understood and turned the three principles into the intelligent cloud's "three service standards"

If Robin Li was responsible for proposing the worldview, then Shen Dou, executive vice president of Baidu Group and president of the Intelligent Cloud Business Group, was responsible for translating it into products—the core commercial landing point of this "intelligent organization" narrative is precisely Baidu's intelligent cloud business.

Every organizational evolution requires a matching set of infrastructure—and the organizational evolution of the AI era requires exactly intelligent cloud computing. In Shen Dou's view, unlike the IT systems of traditional cloud service enterprises, the intelligent cloud must support agents in completing tasks: it must provide computing power, models, and a runtime environment, and must also connect the enterprise's data, tools, and systems, penetrating to the enterprise's underlying resources as never before.

Based on this, Shen Dou clearly stated: Robin Li's three principles of organizational evolution should become the intelligent cloud's three new service standards—if customers are task-centered, then computing power, models, data, and security should be formed into unified infrastructure; if customers want to precipitate tacit knowledge, then help customers preserve and manage organizational context well; if customers want to design closed-loop processes, then help connect task execution, effect evaluation, and feedback optimization.

At the agent level, Shen Dou sorted Baidu's capabilities into three layers: general agents (such as Baidu Dazi, which has opened the Harness engine and Agent API), professional agents (such as Famou for production operations and industrial R&D, whose customers already cover 25 of China's 41 major industrial categories), and customized agents built by enterprises on the Qianfan platform—this platform shares the same origin as Baidu Dazi, also possesses the ability to operate autonomously and continuously evolve, and through dynamic orchestration enables multiple agents to decompose tasks and collaborate in real time.

Several cases were used to demonstrate the gap "from swapping motors to building factories": the lithium battery adhesive company Lanhai Heishi customized three agents—for demand projection, procurement decision-making, and procurement execution—and connected them to its internal OA and ERP. A procurement task that previously required more than ten people from the planning and procurement departments and over 100 work hours each week is now completed by two people in under an hour, and is estimated to bring in several million yuan in profit each quarter; Zheng Yuanyuan, which has nearly 10,000 directly operated stores, uses agents to replace manual supervisor spot checks in pursuit of "one standard across ten thousand stores."

The foundation supporting all of this was also updated at this conference: the sandbox of the Qianfan Agent platform supports millions of concurrent instances and can launch thousands of sandboxes at the hundred-millisecond level; the AI computing platform, for reinforcement learning training, can spin up more than 2 million sandboxes within minutes, with training variance controlled to one ten-thousandth; the Kunlunxin M100 was launched on schedule, and the flagship M300 will be released next year; the newly released Tianchi 1024-card super node adopts optical engine technology, increasing inter-card interconnect bandwidth by 3 times and training performance by 2.5 times, with a single set able to support training of 5-trillion-parameter models and, through architectural upgrades, scalable to 500,000 cards.

He Junjie: Baidu First Acts as Its Own Test Field

In the afternoon session, He Junjie, Senior Vice President of Baidu Group and Head of Group Human Resources and Administration, shared Baidu's practice of integrating the intelligent organization concept into talent development, business collaboration, knowledge management, and process building, adding a "lead by example" link to this narrative—the intelligent organization is not just a concept told to customers; Baidu itself is practicing it internally.

From the talent perspective, the new role of FDE (Forward Deployed Engineer) is "growing out of" the business frontline on its own, connecting the three things of understanding the industry, being able to implement, and knowing how to push products, and has already been delivered and implemented in more than 100 business scenarios across multiple industries in Baidu AI Cloud.

From the organizational perspective, the digital human business department broke down job boundaries to push all employees to transform into Builders, accumulating 130 Skills and 729 organizational assets, with per capita output doubling. The internal AI assistant dodo, built by a team of more than ten people, covers more than 90% of employees and has cumulatively supported tens of millions of conversations.

From the operating mechanism perspective, Looping has expanded from R&D to a broader organizational level—after the network alliance team restructured R&D collaboration into "Agent-driven R&D Looping," the proportion of same-day delivery requirements rose from 7% to 41.9%. In addition, Baidu has established a security governance mechanism covering the full AI lifecycle around five links: identity recognition, permission authorization, data protection, execution supervision, and change management.

Epilogue: A Layout Ten Years Ahead, and Three Predictable Opportunities

Looking back at this conference, Li Yanhong set the tone—the next act of AI is the intelligent organization, and new factories need to be built; Shen Dou took on the implementation—the three principles were translated into three service standards for the intelligent cloud, and three layers of agents plus the Qianfan platform form a complete, sellable toolchain; He Junjie provided the endorsement—Baidu itself has first completed the entire process from embracing artificial intelligence to completing the intelligent organization transformation.

In his speech, Li Yanhong listed three major opportunities of the intelligent organization:

First, asset-light operations. "When one person, or very few people, can lead a group of Agents to complete all kinds of tasks, human resources will no longer be the bottleneck of enterprise expansion, and the pace of iteration will greatly accelerate. For the same matter, an enterprise can even run many small experiments in parallel, rather than betting on only one team at a time. Such an organization can travel light and flexibly explore growth opportunities."

Second, building a new moat. Enterprises can precipitate the tacit experience of top employees into reusable digital assets, then encapsulate them into agent modules that can be deployed globally at millisecond level and run 24/7. Top employees with different skills, after their tacit experience is made explicit, can also learn from each other and achieve mastery through a comprehensive understanding, enabling collective intelligence to surpass individual genius. "In addition, enterprises can, on top of industry-general capabilities, layer on their own unique knowledge, context, culture, and philosophy to form a competitive barrier that is difficult to replicate."

Third, creating and discovering new markets. An organization capable of self-evolution will form a closed loop. It can perceive changes, experiment quickly, learn continuously, and replicate at scale. "Enterprise innovation shifts from relying on a few 'innovation departments' to concurrent exploration across the whole organization, making it more likely to discover and create new markets and cultivate new growth curves."

The blueprint for the new factory has been drawn. The question that follows is: are enterprises willing to pay for rebuilding the factory, and can Baidu's blueprint truly make productivity happen within it? news_keyword_pub,stock,bj920670 iframe below the main text iframe below the main text general statement general statement QR code at the end start QR code at the end start

QR code at the end start QR code at the end start Editor name and work code

Editor name and work code

Source

Sina Tech