11/16 2026
393
Over the past few years, the market has consistently framed AI cloud as a provider of new productivity. Its value has been measured in digital employees, contextual understanding, and safeguards for organizational permissions and data security.
Now, however, at Baidu AI Cloud and across one customer after another, it is becoming clearly perceptible that the value metrics of AI cloud are shifting. On the new battlefield of "intelligent organizations," fine-grained elements such as context, permissions, and data security no longer exist independently. Whoever can combine these capabilities to help enterprises move more quickly from "division of labor and cooperation" to "task units," whoever can better accelerate enterprises' evolution from traditional organizations to intelligent organizations, and whoever can enable intelligent organizations to "grow" on the cloud will earn true B2B value recognition.
From AI productivity to intelligent organizations, a new value battle has begun.
Author | Pi Ye
Produced by | Industry Home
"Building a game used to take at least a month—sometimes even longer—from frontend to backend and from UI to database, and many people had to work together along the way. Now that's completely unnecessary." A game industry practitioner recently told Industry Home excitedly.
Claude Fable 5.2 was officially released recently. With this model, building a game requires only one person and one complete prompt. Within less than 48 hours of the model's release, simulated versions of games such as Honor of Kings and Minecraft were released one after another.
"The organization of the future is a task network." At this year's Baidu World Conference, Robin Li made this assertion.
Behind this assertion, Baidu became one of the first to "turn the knife on itself." One set of data shows that over the past year, Baidu's internal daily active agents (DAA) grew from 5,000 to 37,000, the number of effective tasks completed per week broke through from 84,000 to 1 million, and more than 20,000 organizational skills were accumulated.
Not only Baidu. A real situation is that within more and more large enterprises—such as Industrial and Commercial Bank of China, "Zheng Yuanyuan," and Overseas Sea Group—this model is also quietly emerging. Beyond the existing organizational model, "tasks" are beginning to become new connection nodes, and "people + agents" are beginning to become new execution individuals.
If the surface of organizational circulation relies on people, then the bottom layer relies precisely on systems. Over the past many years, beneath the department-level organizational model grew an IT system spanning from cloud to database to core business software such as ERP, CRM, and MES, and then to OA.
In the current era of intelligent organizations, as agents become new execution units, how should this bottom layer change? Or, to put it another way, in the era of intelligent organizations, what kind of AI cloud foundation do enterprises actually need?
1. Behind Token ROI falling short: "The old continent cannot grow new soil"
"We have decided to wait a few months and observe before making a move." A CIO at a leading chip company told Industry Home.
The real situation of this company is that over the past few years it has made a series of AI attempts—from initially debugging open-source models itself, to later calling mature models and deploying small models locally on its own, to now purchasing a popular office agent on the market. As of this year, after strict internal calculations, "AI has not brought significant business increment."
"Token out of control" was his personal experience.
This is a microcosm, and even not limited to the Chinese market. Recently, McKinsey released a report titled "The State of AI in 2026." The report conducted a real survey on AI implementation among 1,719 enterprise executives across 97 markets. The data shows that 89% of enterprises have already used AI routinely in at least one business function, but only 6% meet the "AI high performance" indicator, almost no growth compared with last year.
Where is the problem?
In fact, the FOMO sentiment and "out of control" phenomenon brought by AI within enterprises have already been discussed in the industry. For example, on Microsoft's official website, it publicly discusses "what the organization of a Frontier Firm should look like"; McKinsey's Agentic Organization report likewise raises a new question, namely: in the AI era, how should enterprises design organizations? The same thing is also happening at Baidu, which focuses on how organizations can grow on their own, continuously evolve, and achieve "measurable + Looping."
A consensus behind these discussions is that the establishment of true AI productivity requires not only bottom-up single-point evolution, but also a top-down new intelligent organization design. More fundamentally, at a time when agents can become execution units and AI large models can penetrate and complete an entire project, a new organizational evolution mechanism is needed to make AI truly an assistant to core business rather than merely an organizational "plug-in."
Objectively speaking, this is not an easy task.
The first issue that bears the brunt is whether agents can become tools of "real productivity." That is, beyond office and Q&A scenarios, in more complex core enterprise scenarios that require reasoning, on factory assembly lines, and in stores, can AI truly become a real "digital employee"?
Second, can organizational knowledge be further accumulated? That is, if past organizational structures deposited experience into the minds of one "old master" after another, then intelligent organizations must extract this knowledge and deposit it at the very bottom of the entire organization, using knowledge bases and model routing as the "enterprise map" for agents to understand and execute.
Finally, and most importantly, this is precisely Baidu's internal exploration: "measurable + Looping." That is, whether tasks are completed. Unlike previous manual acceptance, intelligent organizations need to design different agent acceptance indicators and optimization rules. Based on these rules, agents need to optimize completion by themselves and finally submit for acceptance.
The difficulty lies in the fact that, whether it is task self-execution, organizational accumulability, or AI self-evolution, behind the newer mechanism design, enterprises must simultaneously build a new AI "organizational skeleton." The main body of this skeleton must be agents, a new secure sandbox system, and an AI cloud system built on "Token economics."
Where is the answer?
2. Baidu AI Cloud: A New Positioning in the Era of Intelligent Organizations
Blue Ocean Black Stone is a little giant in China's chemical new materials track. In its daily work, procurement is a particularly "time-consuming and labor-intensive" yet exceptionally important link.
This year, its "procurement system" underwent a completely new evolution: a procurement task that previously required at least 10 people from two teams each week and a cumulative input of 100 work hours can now be completed by fewer than 2 people in less than 1 hour.
The one providing services for it is precisely Baidu Dazi. With the latter's support, Blue Ocean Black Stone built three agents for customized demand reasoning, procurement decision-making, and procurement execution, fully connecting ERP, OA, and WeCom systems. With the support of agents, everyone in the procurement team has the ability to independently complete projects. According to estimates, this alone can bring the enterprise millions of yuan in profit per quarter.
Not only Blue Ocean Black Stone. The same organizational evolution has also appeared within Overseas Sea Group.
As a large hotel group with dozens of hotels and thousands of guest rooms, one high-frequency core task in its daily hotel operation and management is reasonably "adjusting prices" for guest rooms.
In the past, this work was fully handled by operations managers, who had to manually compare prices on OTA platforms every day, estimate customer flow, and finally manually change prices one by one in the system for dozens of hotels. Efficiency and results all depended on personal experience.
Now, the core "executor" of this work is the "room rate management agent" developed on the Baidu Dazi development platform. It fully connects the hotel management system and Ctrip's data, can automatically track market prices every day, find the optimal solution among occupancy rate, room rate, and operating costs, and automatically adjust prices.
It is understood that with the support of this agent, room revenue at Overseas Sea's pilot stores increased by 10%. At the same time, this increase is becoming higher as the agent adjusts and executes more and more times.
A real feeling is that in the scenarios of these enterprises, AI has truly become a tool that accelerates, or even qualitatively transforms, a certain business link. The underlying perspective of this acceleration is that "tasks" become new modular units, and "people + agents" become new execution units.
How did this underlying change happen?
"The three principles of intelligent organization evolution are also the three new standards for intelligent cloud services." At this conference, Shen Dou, Executive Vice President of Baidu Group and President of Baidu AI Cloud Business Group, said, "Organizational evolution and infrastructure development complement each other. Unlike traditional cloud service enterprise IT systems, in the AI era, intelligent cloud services serve specific tasks."
In fact, this precisely constitutes Baidu AI Cloud's comprehensive submission at this conference. More accurately, in the accelerating era of intelligent organizations, Baidu AI Cloud has built in advance a set of "IT frameworks" for advanced enterprises oriented toward the AI era.
First, at the agent level, the "core execution unit," according to different types of tasks within enterprises, Baidu AI Cloud has built an agent system covering general agents, professional agents, and customized agents.
This system takes the general agent Baidu Dazi as its core entry point, based on which enterprises can uniformly invoke and manage all agents. It includes Baidu Dazi for general office scenarios, as well as professional agents such as Famou, Yijian, Yijing, Shengsuan, Miaoda, Kuku, Comate, and Hogee. At the same time, enterprises can build customized agents on their own for their special scenarios and tasks based on the Baidu Dazi development platform, building a "general + professional + customized" agent system suitable for themselves.
The value of this agent system lies in that, as enterprises move toward intelligent organizations, it can better avoid problems such as "silo construction" and "data islands," uniformly and centrally manage all agents through one system, and thereby accumulate implicit knowledge and context systematically and by scenario, better managing and building their own organizational assets.
Second, beneath the agent layer, what Baidu AI Cloud delivers is an infra system centered on "Token" economics, including Baidu Qianfan Agent Infra and Baidu Baige AI Infra. The former supports million-level sandbox concurrency and extremely cost-effective MaaS model services, and at this conference also launched "Ciyuansuo," which, as a neutral and trustworthy large model trading platform, allows enterprises to find Token providers on their own.
The latter, the Baidu Baige platform, can launch 2 million sandboxes in minutes, achieve end-to-end reinforcement learning throughput of 52,000 Tokens per second, and control the training-inference difference within one ten-thousandth.
If the agent system guarantees enterprises' compound interest and value conversion at the agent engineering level, then Baidu AI Cloud's infra system guarantees enterprises' healthy development in Token cost, scenario demand, and security when building intelligent organizations.
That is, through secure sandboxes, extreme MaaS invocation, and reinforcement learning speed, enterprises can obtain underlying guarantees that are safer, more cost-effective, and fully meet demand in overall construction. On the one hand, this guarantee allows enterprises to complete verification of a certain module of an intelligent organization faster and at lower cost, and on the other hand, it also allows them to achieve AI ROI in a more "progressive" way.
In addition, at the conference site, Baidu AI Cloud also released the Tianchi 1024-card super node. This super node adopts the industry's most leading optical engine technology. The G1.5 memory acceleration cabinet can double long-context inference performance, and the G3.5 KV Cache acceleration cabinet, in multi-turn dialogue scenarios with cache hits, reduces first-token waiting time by 90% and increases throughput by 7 times.
From a global perspective, behind this new full-stack AI cloud delivered by Baidu AI Cloud is its underlying thinking about intelligent organizations, namely what kind of agents can execute core tasks well, what kind of unified management can accumulate knowledge assets for enterprises, and what kind of infra system can ensure that intelligent organizations carry out AI transformation at lower cost and higher quality.
Radiated to the product level, these thoughts precisely constitute the new full-stack AI cloud covering "chips, cloud, models, and agents" delivered by Baidu AI Cloud at this conference.
3. Intelligent Organizations Are Already on the Way
In fact, intelligent organizations have already appeared.
Within Baidu, a new type of role named "OPC" emerged this year. This type of role takes tasks as work units and simultaneously completes algorithms, products, solutions, and business sorting. In the past, these things required at least 4 people to complete. Now only one or two people working with agents are needed to run through a customer scenario.
And at Yuanyuan Group, a batch of new "AI store managers" was also officially launched this year. In the past, headquarters needed to send dedicated personnel to each store for inspections and spot checks. Now, with the support of Baidu AI Cloud Yijian, only a camera needs to be installed in the store, and it can serve as a competent AI store manager, monitoring the entire service process and quality and sending reports regularly and on schedule.
Similarly, in large ports and at Haigao Laser, with the support of Baidu Famou, complex tasks that once required specialized algorithm engineers—such as port scheduling, machine cutting, and even chip EDA simulation verification—are no longer handled exclusively by them. Now, a single OPC role, combined with Baidu Famou's 'self-evolution' capability, can deliver optimal solutions for specific production links and execute specific tasks more efficiently.
The one providing intelligent organizational support for these enterprises is none other than Baidu AI Cloud. It can even be said that the suite of capabilities offered by Baidu AI Cloud—from MaaS to agents, from Baidu Dazi to Baidu Famou and Yijian, and further to the Baidu Dazi development platform and Baidu Dazi open platform—not only supports the organizational evolution of enterprises at a higher level but also provides targeted acceleration for the intelligent evolution of organizations.
This acceleration lies in the fact that, at a time when AI is speeding up business implementation and reshaping the operating models of enterprise business, enterprises can leverage this sufficiently 'handy' and intelligent tool to integrate AI into their organizations with less resistance and complete the transformation from traditional organizations to intelligent organizations.
Over the past few years, the market has consistently defined AI cloud as a provider of new productivity. Its value has been measured in terms of digital employees, contextual understanding, and safeguards for organizational permissions and data security.
Now, however, within Baidu AI Cloud and across one customer after another, it is clearly perceptible that the value metrics of AI cloud are changing. In the new battlefield of 'intelligent organizations,' fine-grained elements such as context, permissions, and data security no longer exist independently. Whoever can combine these capabilities to help enterprises more quickly achieve execution—from 'division of labor and cooperation' to 'task units'—whoever can better provide enterprises with acceleration in the evolution from traditional organizations to intelligent organizations, and whoever can enable intelligent organizations to 'grow' on the cloud will gain true B2B value recognition.
From AI productivity to intelligent organizations, a new value battle has begun.