EN / 中文

Beyond AI Chips and Models: Why Intelligent Organization Is the Next Battleground

by guangzixingqiu·November 14, 2026

by | Wu Xianzhi

In 2016, the main narrative for domestic tech companies was the Internet, while the theme of Baidu's annual conference that year was "AI is the new Electricity."

At the conference, Baidu unveiled a series of new AI products and technologies for the first time. These developments either laid the core foundation for Baidu's AI efforts in the future or became the technical cornerstone for businesses such as Xiaodu and Luobo Kuaipao.

"What is the next act after mobile internet? It is actually quite clear now—it is what we call Artificial Intelligence (AI)." This counter-consensus prediction made by Robin Li at the time has become common knowledge today.

Today, with the theme of Baidu World 2026 being "The Next Act: Intelligent Organization," Robin Li, who already had the answer ten years ago, offered a new prediction.

"If AI is the new electricity, then the Intelligent Organization is the new factory of the AI era—the place where productivity truly happens. Only when organizational structures matching various new AI capabilities are implemented will productivity see a substantial increase."

He believes that as AI technology is deployed on a large scale, the entire industry needs to shift its focus to organization. To this end, he proposed three principles for Intelligent Organizations: breaking down job boundaries, accumulating tacit knowledge, and designing closed-loop processes.

Currently, Baidu has completed its full-stack system of "Chip, Cloud, Model, and Agent" and continues to deliver various AI products to Government (G), Business (B), and Consumer (C) sectors. Robin Li's speech at this juncture clearly reflects his awareness that the Intelligent Organization is not only Baidu's next internal step but also the industry's next destination amidst the AI wave.

If You Change the Motor, You Must Build a New Factory

The outbreak of a large-scale productivity revolution and the realization of its value often do not happen simultaneously, because technology arrives first, while the adapted organizational systems have yet to emerge.

In the 1770s, improved steam engines appeared, meaning production sites no longer required rivers as a rigid criterion for location selection.

Managers at the time believed that connecting steam engines to existing workshop equipment would allow craftsmen to continue producing using traditional methods. Although power shifted from natural water energy to steam power, the organization remained stuck in the past. Consequently, machines did not rapidly improve production efficiency as expected.

It was not until half a century later, with the establishment of the modern factory system featuring process segmentation, fixed workstations, fixed shifts, and unified management, that technology truly delivered its productivity value.

Subsequent technological revolutions driven by energy generally experienced similar cycles.

New electric motors entered factories, but the corresponding organizational forms waited until Ford's Model T assembly line in 1913. This organizational system, which organized distributed electric drive and sequenced processes, broke the bottlenecks constraining productivity. The emergence of new motors preceded new factories, and the same pattern held true in the Internet era.

From informatization construction to the subsequent digital revolution, enterprises launched one system after another, but the corresponding organizational capabilities, such as OKR frameworks, arrived belatedly only in the last decade.

Now, with the surge in AI applications, the weak-link effect between technology and organization has emerged once again. Model capabilities undergo generational changes; in just two years, agents have evolved from conversational bots to long-chain task executors. On the enterprise side, however, since organizational systems adapted to AI have not kept up, the curve of corporate capital expenditure has not risen in sync with the curve of revenue growth.

In its "2026 Work Trend Index Annual Report," Microsoft analyzed the implementation of enterprise AI over the past few years and discovered a paradox: employees began using AI tools to boost personal efficiency, and teams fully embraced new tools, yet business processes, collaboration methods, and evaluation mechanisms remained stuck in the pre-AI era. The organization itself underwent no change.

Improving local efficiency does not mean organizational productivity is directly related to the number of AI tools. Tool deployment and pilot results can easily be framed as impressive performance reports, leading people to mistakenly believe that transformation is complete.

However, in enterprise practice, the time saved by employees using AI is ultimately consumed by cross-departmental alignment, multi-layer approvals, and repeated reviews within organizational processes. This results in increasingly long token bills without a corresponding rise in profit curves. The primary reason lies in processes, collaboration, and evaluations—three areas that still reflect the conditions of the previous era.

Boston Consulting Group (BCG) highlighted a typical case study of a European energy company in an article titled "Design Your Company for AI, Not AI for Your Company."

This company initially conducted numerous AI pilots. Even though individual tests succeeded, overall benefits were insignificant. Consequently, the enterprise abandoned the strategy of changing only tools without altering institutions, shifting instead to restructuring customer processes, which marked the turning point for business improvement.

Similar cases exist domestically. Last year, many large companies pursued comprehensive AI adoption through a "rapid expansion" narrative, including unlimited token supply. However, the situation changed drastically this year.

An executive from one giant summarized the issues in August as four mismatches: cognitively, enterprises either exaggerated or underestimated AI; in terms of efficiency, large models were used for daily tasks without touching core functions; regarding scenarios, AI remained on the periphery, failing to integrate into core operations; and in assessment, there was a pursuit of immediate results while ignoring internal adaptability.

Local success does not equal organizational enhancement. In his speech, Robin Li stated that no company becomes more competitive overnight simply by using a few AI tools.

AI is the new motor providing kinetic energy, and the Intelligent Organization is the accompanying new factory. Motors can be procured internally or externally, but only the factory requires customized planning by the enterprise itself. Since the motor has been changed, the factory must also be replaced.

Organization: The Next Battleground in AI Competition

Many enterprise managers believe that deploying more AI tools will lead to victory in the AI race.

This view is largely influenced by the overseas theory of the "Five Layers of the AI Cake." This framework decomposes the AI industry into the Energy Layer, Chip Layer, Infrastructure Layer, Model Layer, and Application Layer, describing them as interconnected and linked upstream and downstream. This framework essentially answers "Where does intelligence come from?" and focuses on energy competition.

Some argue that the essence of AI competition is an energy game, which fails to address how AI truly becomes productive force. At Baidu World 2026, Robin Li argued that the next act of AI lies in the Intelligent Organization. Taking Baidu as an example, beyond the full-stack foundation of "Chip, Cloud, Model, and Agent," an "Organization Layer" needs to be added, focusing on how intelligence, once generated, truly enters enterprises and creates productivity.

Following this logic and linking it to Robin Li's previous statements, one can completely understand the evolution of the "Intelligent Organization" along a vertical timeline.

As early as the 2025 Baidu World Conference, Robin Li attempted to shift attention from a technical perspective back to an organizational one. When discussing the internalization of AI capabilities, he pointed out that true internalization lands on the organization, not the individual. The core is transforming the business problems an enterprise needs to solve into problems that AI can undertake and process.

By May of this year, building on the internalization of AI, he posited that enterprises in the AI era should move toward organizational "Self-Evolution." Self-evolution has three dimensions: agents need to evolve from passive response to active execution; human individuals need to learn to coexist with AI, evolving from ordinary individuals into super-individuals; and enterprise organizations must transform division of labor among humans into mixed squads of humans and agents, becoming super-organizations.

Along this line of thought, Baidu World 2026 landed on "Intelligent Organization," stating: "Organization is an inevitable proposition in the AI era; the Intelligent Organization is the factory that activates productivity in the AI era." In the second half of AI competition, whoever solves the difficulty of organizational implementation is the one who can truly convert model capabilities into industrial dividends.

Meanwhile, with the explosion of agents, Robin Li provided a metric called DAA (Daily Active Agents) at the Baidu Create Conference in May to measure AI delivery results.

First, internalize AI into organizations and individuals; then promote active evolution of both; use DAA as a ruler; and finally move towards the Intelligent Organization.

Traditional organization was built upon the job position system, deriving authority, responsibility, processes, and systems from individual capabilities. The logic of the Intelligent Organization shifts to a task-centric model, allowing people and agents within the organization to collaborate around goals.

Robin Li first constructed a foundational framework for the Intelligent Organization, namely the Three Principles of Intelligent Organization. The first is breaking down job boundaries to release organizational vitality; the second is accumulating tacit knowledge to build organizational intelligence; the third is designing closed-loop processes to drive organizational evolution.

These three principles progress layer by layer, from releasing individual efficacy, to aggregating collective intelligence, to forming optimization loops, ultimately constituting the skeleton of the Intelligent Organization. Within this, an evolution formula of "Measurable + Looping = Continuous Evolution" forms the optimization loop.

Only when the organization's collaboration methods, knowledge management, and iteration mechanisms achieve intelligent adaptation can the new motor and new factory unleash organizational efficiency in the AI era.

The First Batch of "Intelligent Organizations" Has Arrived

Amidst the AI wave, different institutions have offered solutions for organizational efficiency gains in the AI era based on their own characteristics.

Microsoft's "Frontier Firm" emphasizes organizational form; McKinsey's "Agentic Organization" focuses on organizational methodology; while Anthropic's advocated "AI Organizations" prioritize human-machine team configurations. Robin Li proposed the "Intelligent Organization," focusing on how the organization grows itself.

Relying on its full-stack self-developed capabilities in "Chip, Cloud, Model, and Agent," Baidu has established unified quantitative assessment, task feedback, and looping closed-loops, achieving practical results internally.

In implementation within Baidu, the "DuMate" team took the lead in consolidating the collaboration chain among product, design, and R&D into a reusable human-machine collaborative workflow.

In this process, product managers input scenarios and judge directions; agents undertake requirements analysis, generating high-fidelity prototypes and implementation plans according to frontend specifications, and adjusting experience and interaction with designers; developers receive complete prototypes and PRDs (Product Requirement Documents) to develop directly, finally completing walkthroughs against specifications.

Throughout the entire workflow, product managers focus more on judgment, designers on experience innovation, and developers reduce the cost of understanding requirements. A typical requirement cycle from analysis to deliverable solution shortened from 7 days to 2 days. This entire process has been consolidated into a "Requirements Closed-Loop Workflow" Skill. Other product teams can reuse it once they configure product information, code repositories, and requirement sources.

Over the past six months, the team and DuMate together completed thousands of tasks. Experience from each task carries forward to the next, manifesting as self-iteration of the product. In half a year since launch, the version underwent 160 iterations. Team members noted: "Often, so-called job boundaries are not entirely boundaries of capability itself, but boundaries of knowledge and experience flow."

This paradigm of Intelligent Organization, capable of self-iteration and continuous accumulation, diffuses outward from full-stack AI to industries.

In the power O&M scenario, DuMate entered China Southern Power Grid Dispatch Center (CSG Dispatch), jointly creating the product "Tianxuan Partner."

Facing the heavy responsibility of ensuring electricity supply for over 200 million people across five southern provinces, distribution network self-healing lines could quickly locate faults and restore power. However, Guangxi Power Grid alone experienced 8,215 permanent faults in 2025. Traditional manual judgment took over 60 minutes per instance, heavily relied on expert experience, and was prone to misjudgment.

To solve this problem, Tianxuan Partner was deployed at CSG Dispatch. By learning from tens of thousands of historical fault handling records, it accumulated senior expert experience into reusable professional knowledge. It linked fault records, switch signals, and line topology into a complete evidence chain, achieving closed-loop judgment of "Fault-One Analysis" with an accuracy rate exceeding 70%.

It is understood that this application improved distribution network fault analysis efficiency by 12 times and has currently been fully promoted across 14 prefecture-level cities and 112 counties in Guangxi. Its "accumulated tacit knowledge" allows veteran experts' practical experience to be passed down, with wisdom continuously evolving during every operation.

Since the same product generates results both internally and externally, the concept of "New Motor + New Factory" has found its best proof. We also note that changes within and outside Baidu provide a new hint: moving towards Intelligent Organization is a "Top Leader Project."

The current explosion of AI applications validates Robin Li's judgment from ten years ago; the top leader's correct directional judgment is the prerequisite for obtaining a ticket to the next era. Now, the implementation of Intelligent Organization requires the top leader to personally engage.

Baidu and the first batch of Intelligent Organizations have already achieved results. The second and third batches will follow.