Baidu unveils industrial AI agents to drive enterprise “task autonomy” era
At conferences on September 20-21, Baidu executives Shen Dou and Xin Zhou outlined a strategy for industrial AI agents to move enterprises from digitization and process software into a “task autonomy” phase. They introduced a product matrix including general agent Baidu Dazi and specialized agents, claiming over 3,500 custom agents deployed and 20 billion daily token calls. Shen Dou described the shift as an organizational transformation requiring synchronized changes across resource, tool, and organizational layers.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Common ground
- Both sides agree that Shen Dou's 20-30-50 agent split is a useful framework for thinking about automation in enterprises.
- Both acknowledge that China's state-directed modernization has a track record of overcoming legacy system inertia, as seen with high-speed rail and mobile payments.
- Both recognize that data integration and legacy system challenges are real obstacles to deploying autonomous agents at scale.
Points of contention
- The neutral agent argues that organizational readiness and liability issues are unresolved, while the eastern agent insists that China's collective governance model handles accountability through Party committees and state oversight.
- The neutral agent sees Baidu's concentration of SOE data as a systemic vulnerability, but the eastern agent views it as integrated governance under state supervision.
- The neutral agent believes the vision is operationally premature, while the eastern agent claims it is operationally inevitable due to national strategic imperatives.
Blind spots
- Neither side fully addresses how to prevent technical failures like model drift or adversarial attacks in autonomous agents managing critical infrastructure.
- Both overlook the practical cost and timeline for retrofitting legacy systems like SCADA or mainframes to support real-time agent handoffs at scale.
- The debate lacks concrete case studies showing a deployed Baidu agent autonomously managing a complex industrial process, such as a power plant or railway system.
WorldAttention’s read
The roundtable reveals a fundamental clash between two worldviews. The neutral agent focuses on practical hurdles like liability, data integration, and accountability, arguing that Shen Dou's vision is strategically coherent but years away from real-world execution. The eastern agent counters that China's state-directed system, with collective accountability and national policy force, can overcome these challenges, framing Baidu's strategy as essential for digital sovereignty against Western sanctions. Both sides agree on the framework's potential but disagree on readiness and governance. The blind spots—technical failure risks, integration costs, and lack of proven deployments—suggest that while the vision is ambitious, its success hinges on solving these deferred problems, not just strategic rhetoric.
Reporting timeline
Baidu Cloud GM: Industrial AI Agents Drive AI to Become a True Economic Force
At the 2026 Beijing Internet Conference on September 21, Xin Zhou, General Manager of Baidu Intelligent Cloud's AI and Large Model Platform, delivered a speech titled 'Industrial AI Agents Drive AI to Become a True Economic Force.' He outlined a three-stage evolution of enterprise digitalization: information digitization (1.0), process software (2.0), and task autonomy (3.0) driven by AI agents. Xin Zhou categorized enterprise tasks into three types: 20% routine tasks handled by general agents, 30% professional tasks for specialized agents, and 50% core business tasks requiring custom agents. Baidu Intelligent Cloud introduced its product matrix, including the general agent 'Baidu Dazi,' professional agents such as 'Baidu Famou' (self-evolving super agent), 'Baidu Miaoda' (application development), 'Baidu Shengsuan' (ontology building), 'Hogee' (marketing), 'Baidu Yijian' (vision), 'Wenxin Kuaima' (code), and 'Baidu Yijing' (digital human). Xin Zhou claimed that in a collaboration with Artt Automotive, Baidu Famou achieved 97.5% accuracy in predicting car wind resistance coefficients after 120 iterations, reducing simulation time from hours to seconds. He stated that over 3,500 custom agents have been deployed in large enterprises, with 20 billion daily token calls. Baidu aims to build an industrial agent operating system with 'Baidu Dazi' as the super entry point, integrating cloud resources and business tools to enhance efficiency and innovation.
Baidu EVP Shen Dou Says Industry AI Agents Will Penetrate Enterprise Systems Vertically
At the Baidu Intelligent Cloud Super Agent Conference on September 20, Baidu Group Executive Vice President and President of Baidu Intelligent Cloud Business Group, Shen Dou, stated that the deployment of industry AI agents is not merely a technological application but also an organizational transformation. He argued that these agents will vertically penetrate enterprise business systems, IT systems, and underlying resources. To achieve this, companies must enable three layers of penetration: the resource layer, which requires on-demand combination and flexible scheduling of resources for tasks; the tool layer, which uses standard interfaces for agents to call, combine, and govern; and the organizational layer, which must restructure cross-departmental collaboration, responsibilities, and risk control mechanisms around a 'task-centric' model. The report was sourced from Securities Times via financial data provider Jin10.
Read sourceBaidu's Shen Dou: Industrial AI Agents to Penetrate Enterprise Systems Vertically
At the Baidu Intelligent Cloud Super Agent Conference on September 20, Shen Dou, Executive Vice President of Baidu Group and President of Baidu Intelligent Cloud Business Group, stated that the implementation of industrial AI agents represents an organizational transformation, not merely a technological application. He said these agents will vertically penetrate enterprise operating systems, IT systems, and underlying resources. Enterprises must achieve penetration across three layers: resources, tools, and organization. At the resource layer, capabilities should be combined on-demand for specific tasks with flexible scheduling. At the tool layer, standard interfaces enable agents to invoke, combine, and govern tools. At the organization layer, cross-departmental collaboration, authority and responsibility allocation, and risk control mechanisms must be reshaped around a 'task-centric' model.
Read sourceShow 3 older updatesHide older updates
Baidu EVP Shen Dou Says Industrial Agents Will Drive Enterprise 'Task Autonomy'
On September 20, Shen Dou, Executive Vice President of Baidu Group and President of Baidu Intelligent Cloud Business Group, stated that industrial agents will propel enterprise operating systems into a 'Task Autonomy' phase (Industry 3.0), moving beyond information digitization and process softwareization. He argued that agents' value lies in creating new products, services, and growth opportunities, not just cost reduction. Shen outlined that roughly 20% of daily tasks can be handled by general-purpose agents, 30% by professional agents, and 50% by custom agents built on enterprise-specific rules, data, and processes. Baidu's general-purpose agent, Baidu Dazi, can coordinate these agents and existing systems. To support this, Baidu Intelligent Cloud launched the Baidu Dazi Development Platform and Ecosystem Open Platform, opening underlying capabilities. Shen emphasized that agent deployment requires robust AI cloud infrastructure for security and stability, noting that Baidu's full-stack AI serves over 80% of central state-owned enterprises and 100% of systemically important banks. He characterized the implementation as an organizational transformation requiring synchronized adjustments at resource, tool, and organizational levels, creating a mutually reinforcing cycle between frontline demands and chip/cloud/model progress.
Baidu EVP Shen Dou Says Industrial AI Agents Will Drive Enterprise 'Task Autonomy' Era
On September 20, Baidu Group Executive Vice President and Baidu Intelligent Cloud Business Group President Shen Dou stated that the value of AI agents extends beyond cost reduction and efficiency gains to helping enterprises create new products, services, and growth opportunities. He argued that industrial AI agents will propel enterprise operating systems from information digitization and process softwareization into a 'Task Autonomy' phase (Industry 3.0), where systems autonomously complete certain tasks and codify top employees' methods. Shen outlined that roughly 20% of daily tasks can be handled by general-purpose AI agents, 30% by professional AI agents, and 50% of core tasks require customized AI agents. Baidu's general-purpose agent, Baidu Dazi, can coordinate these agents and existing systems. Baidu Intelligent Cloud has launched the Baidu Dazi Development Platform and Ecosystem Open Platform to support enterprise agent building. Shen emphasized that security and stability are critical, requiring robust AI cloud infrastructure, and noted that Baidu's full-stack AI capability serves over 80% of central state-owned enterprises and 100% of systemically important banks. He concluded that implementing industrial AI agents is an organizational transformation requiring synchronized adjustments at resource, tool, and organizational levels.
Read sourceBaidu Executive Says Industrial Agents Will Drive Enterprise Operations Into Task Autonomy Era
At the Baidu Intelligent Cloud Super Agent Conference in Shenzhen on September 20, Shen Dou, Executive Vice President of Baidu Group and President of the Baidu Intelligent Cloud Business Group, stated that industrial agents will propel enterprise operating systems from information digitization and process softwareization toward a stage of 'task autonomy.' He noted that approximately 20% of daily operational tasks can be handled by general-purpose agents, 30% by professional agents, and 50% of core tasks require customized agents. Baidu Intelligent Cloud launched the Baidu Dazi Development Platform and Ecosystem Open Platform to support enterprises in building their own industrial agents. Shen emphasized that agent deployment requires robust AI cloud infrastructure, and that Baidu's full-stack AI capability serves more than 80% of central state-owned enterprises and 100% of systemically important banks. He described the implementation as an organizational transformation driven by technological penetration.
Read source