Li Yanhong Proposes the Intelligent Organization at Baidu World 2026, Says the Next Act of AI Is Reconstructing Organizations
At the Baidu World 2026 conference on November 12, Li Yanhong proposed the intelligent organization, arguing that AI has not produced the expected economic benefits because enterprises have only replaced the motor without redesigning the factory. The conference disclosed that the number of daily active agents inside Baidu rose from an average of 5,000 per day to 37,000 over the past year, that the effective tasks completed by agents each week increased from 84,000 to more than 1 million, and that 4,300 Workspaces have been built and more than 20,000 Skills accumulated. The article also reviews his successive judgments from "AI is the New Electricity" in 2016 to DAA in 2026, and lists external cases such as China Southern Power Grid and Haigao Laser.
Why read this
AI score 53
The article sorts through the timeline of Li Yanhong's judgments over ten years and presents data on agent operations inside Baidu, which can serve as background for understanding the notion of the intelligent organization.
Topics
Body · AI translation
"Tech prophet" Robin Li: Does he see AI's next decade? New Weekly Download the app for exclusive early access November 13, 2026, 17:58:16 From Guangdong Province
Over the past decade, how many "next acts" has Robin Li actually called correctly?
What is the basis for the "intelligent organization" he proposed? And if he is right again this time, what Chinese companies will face next is not merely a technological upgrade, but a redesign of the organization itself.
At the Baidu World conference ten years ago, the theme on stage was just one sentence: "AI is the New Electricity."
That year was 2016. Robin Li stood on stage and said that the next act of the mobile internet is artificial intelligence. Some in the audience applauded; others privately wondered: a search company says it wants to become an AI company—where would that story even begin?
In May this year, when everyone was eager to talk about how many Tokens were consumed each day, Robin Li proposed a new yardstick for measuring AI value at the Create conference—DAA (daily active agents). Afterward, the industry also began to re-examine the practice of measuring AI value solely by Token consumption.
This time, Robin Li spoke again. On November 12, at the "Baidu World 2026" conference (hereinafter referred to as "the conference"), he pointed the "next act" toward a term that does not sound very technical—the intelligent organization.
A comparison of Baidu founder and CEO Robin Li's "next acts" over ten years.
Over the past decade, how many "next acts" has Robin Li actually called correctly? What is the basis for the "intelligent organization" he proposed? And if he is right again this time, what Chinese companies will face next is not merely a technological upgrade, but a redesign of the organization itself.
On November 12, 2026, the "Baidu World 2026" conference and exhibition was held at the China National Convention Center in Beijing.
The "next acts" he announced in advance in those years
Robin Li always likes to state his judgments affirmatively before things happen, and every time they stand the test of time. This seems to have become his habit, so much so that netizens happily call him a "tech prophet."
In the summer of 2017, he rode in a self-driving car onto Beijing's Fifth Ring Road and got a ticket. But the angle from which he responded at the time was strikingly original: "If the ticket for self-driving has already arrived, can mass production of self-driving cars be far behind?"
A ticket now looks like a small footnote to an event. But at the time, it meant Baidu had pushed self-driving from the laboratory onto real roads. At almost the same time, Waymo on the other side of the ocean also launched early passenger tests in Phoenix. Tech companies in China and the United States were doing almost the same thing in sync: teaching machines to move in the real world.
In 2018, Baidu released China's first full-featured cloud AI chip. Robin Li's judgment at the time was also simple: once AI applications explode, demand for algorithms and computing power will grow exponentially. In the same year, Amazon AWS launched its first self-developed CPU. Global tech giants coincidentally deepened their layouts in underlying computing power, like a convergence with no prior agreement.
In 2022, Baidu World showed an image: the ERNIE large model "completed" the painting Dwelling in the Fuchun Mountains. Robin Li said that AIGC would become a new content production method after PGC and UGC. Four months later, ChatGPT was released, and generative AI swept the world.
In 2023, Baidu released its first generative language large-model product. Robin Li did not stop at the narrative of "how strong the model is." He said something that sounded a bit ahead of its time: redo all products with large models. At almost the same time, Microsoft launched Microsoft 365 Copilot, embedding AI into office software. The value of large models began to shift from "capability demonstration" to "application reconstruction."
At Baidu World 2024, Robin Li demonstrated on site a small team composed of five agents—team leader, planner, editor, programmer, and quality inspector—that built a conference registration system in a few minutes. He said that agents are the most mainstream form of AI applications. Half a year later, the American AI company Anthropic disclosed the Research multi-agent architecture. The first year of agents began, and division of labor and collaboration among multiple agents became one of its most prominent features.
By 2026, Robin Li proposed DAA (daily active agents), saying it is the metric that truly corresponds to DAU in the mobile internet era—that is, focusing on how many agents are continuously working for people and delivering results every day. Since then, the measurement of AI value has also shifted from "how many Tokens were consumed" to "what tasks were completed."
If we line up the nodes mentioned above, we may discover a pattern: Robin Li's judgments are almost always earlier than industry consensus. From chips to models, from models to applications, from applications to agents. Baidu has not done everything best, but it has never changed its main direction of AI.
From 2016 to 2026 is exactly the decade in which artificial intelligence technology formed an independent tech stack. If the "chip-cloud-model-agent" full stack that Baidu spent ten years building (Kunlun Chip-Baidu AI Cloud-ERNIE large model-agent applications) is a set of "electric motors," then now he senses that the future competitiveness of enterprises will not lie in how many "electric motors" they own, but in whether they can grow a new form that organizes these "electric motors"—a new "factory building."
Panoramic view of the Baidu World 2026 conference.
The narrative of "organization" is quietly changing.
The term "intelligent organization" sounds like a management academic term. But the question it seeks to answer is actually very old—when a group of people work together, how exactly should they be organized?
About 170 years ago, the New York and Erie Railroad Company drew a chart. It was a tree-like organizational chart, with distinct branches, each labeled with functional departments and personnel assignments.
One of the earliest organizational charts in the world.
What they wanted to solve at the time was simple: when a company grows too large for one person to manage, how do you keep it running in an orderly way? Later, sociologist Max Weber systematized this logic into "bureaucracy"—professional division of labor, clear authority and responsibility, hierarchical management, and institutionalized processes. Large-scale production in the industrial age relied on this skeleton.
In the internet age, Amazon's Bezos proposed another metaphor: the two-pizza team. A team should be small enough that two pizzas can feed everyone. Control team size, delegate decision-making authority, and reduce hierarchy and coordination costs. Behind the two pizzas is the internet age's pursuit of "agility"—faster response to the market, less bureaucratic waste.
These two stages share one thing in common: the participants in the organization are always only people. All structural design is answering "how to organize people more efficiently."
The AI era has changed this premise for the first time. Organizations now have a new kind of member: agents. They do not need workstations, do not get tired, and can be copied ten thousand times in the same second to execute ten thousand different tasks. The question is no longer "how to manage people," but how to organize people and agents at the same time, so that together they create greater value than they could alone.
Baidu's digital human "Xijiajia."
Baidu is not the only one thinking about this. In 2025, McKinsey proposed the concept of the "Agentic Organization," calling it the biggest organizational paradigm shift since the industrial and digital revolutions. Microsoft, in its 2025 Work Trend Index, proposed the "Frontier Firm," describing an organizational form built around on-demand intelligence and "human + AI agent" hybrid teams. The brightest minds in the world are searching for answers in the same waters.
He has a metaphor that has been repeatedly cited: AI is the new electricity. But electricity itself does not create value. Install a new motor in an old factory, and if the conveyor belt has not changed, the process has not changed, and the workers' positions have not changed, the efficiency gain is limited. What truly changes productivity is redesigning the entire factory. And the intelligent organization is the new factory building he speaks of.
On the stage of Baidu World 2026, he offered a more specific diagnosis: "AI has not produced the expected economic benefits because we have only replaced the motor without redesigning the factory. No company will become more competitive overnight just because it uses a few AI tools." He believes that only when an enterprise can restructure its organization around AI capabilities can it seize the enormous opportunity for success that this era offers.
From "motor" to "factory building,"
Baidu first experiments on itself
Proposing a concept is easy. What is hard is finding a place willing to let you take it apart and rebuild it. Baidu's choice was to first run an intelligent organization experiment on itself.
The conference disclosed a set of data: over the past year, Baidu's internal daily active agents (DAA) grew from an average of 5,000 per day to 37,000, more than a 6-fold increase; the number of valid tasks completed by agents grew from an average of 84,000 per week to more than 1 million per week, an 11-fold increase. Today, Baidu has built 4,300 Workspaces, with more than 20,000 Skills accumulated across the organization.
"Let one person's experience become the capability of the entire organization," Li Yanhong said. For a technology company with tens of thousands of employees, organizational inertia is enormous. Hierarchy, processes, departmental walls, reporting lines—these things are deeply entrenched in any large company. But some changes are taking place inside Baidu.
The Baidu Dazi team is a good example. It iterated more than 160 times in half a year, with 90% of the code and 50% of external communication materials generated by Dazi. One speaker on the team was doing quantum computing research in a physics lab half a year ago. After joining Baidu, with the help of Baidu Dazi, this person simultaneously took on three roles—product, R&D, and project management—advancing from analyzing problems and forming solutions all the way to verifying results. Over the past six months, the team and agents together completed tens of thousands of tasks, and the experience from each task was carried into the next. This is not one person's capability suddenly broadening; it is the organization's "way of knowledge flowing" that has changed.
At this conference, another capability repeatedly mentioned was Looping. Li Yanhong gave a formula: Measurable + Looping = Continuous evolution. "The company becomes a highly automated operating organization composed of countless Loops. The organization becomes a system that continuously learns and evolves."
Baidu has built full-stack capabilities around "chip-cloud-model-agent," but what Li Yanhong values is not how strong each of these four layers is individually, but whether they can form a cycle—chips support the cloud, the cloud supports models, models support agents, and agent feedback returns to the optimization of chips and the cloud. A closed loop, getting faster with every cycle.
Leaving Baidu, this logic began to be tested in real industries. China Southern Power Grid's Tianxuan partner learned from tens of thousands of historical fault-handling records, distilling the experience of veteran workers into reusable knowledge; its distribution fault analysis accuracy has reached over 70%, with efficiency improved 12-fold, and it has been rolled out in 14 prefecture-level cities and 112 counties in Guangxi.
Haigao Laser's multi-robotic-arm scheduling used to rely on a handful of process experts allocating tasks based on experience; Baidu Famou turned that experience into algorithmic capability that the entire production team can reuse. These cases seem to share a common structure: first precipitate tacit knowledge, then let knowledge flow and evolve in every task.
This corresponds exactly to what Li Yanhong calls the "three principles of the intelligent organization"—break down job boundaries and unleash organizational vitality; precipitate tacit knowledge and build organizational intelligence; design closed-loop processes and drive continuous evolution. These three things are not parallel; they are progressive. Without the first, knowledge will not flow; without the second, the flow has no content; without the third, the flow will not become smarter.
Understand the intelligent organization in one chart.
It is understood that at this year's conference Baidu also released a series of products around the four-layer architecture of "chip-cloud-model-agent." The large-model trading platform "Ciyuansuo" was officially launched; Wenxin 5.5 Flash Preview was officially released; the Wenxin Lightning deep-thinking model and the Wenxin Yijing micro-expression large model simultaneously went live on the Qianfan platform.
(Image/Baidu World 2026)
Baidu Yijing launched "Short Comic Drama Studio"; the video clip of the digital human Xijiajia interviewing Li Yanhong shown at the conference was generated by Baidu Yijing; Baidu Dazi was upgraded to a unified entry point connecting all professional agents; Baidu Famou was upgraded to 3.0; Kunlun Chip M100 has also been listed on schedule, among others; in addition, Apollo Go also covers 28 cities worldwide, with autonomous driving mileage exceeding 350 million kilometers and fully driverless mileage exceeding 240 million kilometers, and has partnered with CAR Inc. to launch the world's only driverless-car "road trip" service model.
Xijiajia digital human interviews Robin.
Li Yanhong believes that the future "intelligent organization" will embrace three major opportunities. First, asset-light operations. Very few people can lead a group of Agents to complete all kinds of tasks; second, build a new moat. Enterprises can precipitate the tacit experience of top employees into reusable digital assets; third, create and discover new markets. 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.
He once said something that may explain why it is again he who first speaks of the "next act."
"We believe in technology, not since three years ago, not even since ten years ago, but from the very first day Baidu was founded." Ten years ago, this sentence sounded like self-persuasion. Today, it reads more like an after-the-fact annotation.
Whether the intelligent organization is AI's "next act," no one can answer right now, but one thing is certain. When companies around the world are asking "why hasn't AI investment translated into growth," someone has pulled the question back from the technical level to the organizational level. This turn itself has already gone one step further than most discussions about AI.
"Special statement: The above content (including videos, pictures or audio, if any) is uploaded and published by users of "Dafeng Hao," a self-media platform under Phoenix Net. This platform only provides information storage space services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."