Baidu Intelligent Cloud GM Xin Zhou: Industrial AI agents drive AI to become a true economic force

Beijing, September 21 — At the "2026 (23rd) Beijing Internet Conference," Baidu Intelligent Cloud's General Manager of AI and Large Model Platform, Xin Zhou, delivered a keynote speech titled "Industry AI Agents: Driving AI to Become a True Economic Force." The address focused on the core direction of industry AI agents, outlining the pathway for AI to evolve from a technological capability into tangible industrial economic value, and sharing Baidu Intelligent Cloud's latest developments in this domain.

Industry AI Agents Propel Enterprises into the 3.0 Era of Task Autonomy

According to Xin Zhou, enterprise digitalization has progressed through three distinct stages:

• 1.0 – Information Digitization: Leveraging databases and Office tools to migrate offline information online. The core focus was on memory and governance.
• 2.0 – Process Softwareization: Using ERP, CRM, OA, and other systems to standardize business processes. The core focus was on process governance.

The essence of the first two stages was to enhance individual human capabilities and improve collaboration efficiency.

With the arrival of the 3.0 era, led by industry AI agents, enterprise operations enter a phase of task autonomy and task governance. AI agents, powered by reasoning and long-chain thinking capabilities, can automatically capture, iterate, and upgrade human knowledge and experience, thereby raising the average capability of overall business operations.

Xin Zhou categorized enterprise tasks into three types:

• 20% – Routine operational tasks: Handled by general-purpose AI agents.
• 30% – Professional domain tasks: Managed by specialized AI agents.
• 50% – Core enterprise tasks: Deeply tied to a company's unique business and internal experience, requiring customized AI agents.

These three types of agents are not isolated. Specialized and customized agents can be invoked by general-purpose agents via skill components, creating a unified business entry point. This frees employees from琐碎 tasks, allowing them to focus on core business innovation and decision-making.

Xin Zhou introduced Baidu DaZi, a general-purpose AI agent launched by Baidu Intelligent Cloud. It is built on the Harness system and a three-layer memory architecture, integrating a rich plugin ecosystem from Baidu Search and Baidu Wenku. It supports both personal and enterprise editions, as well as multiple deployment modes including public cloud, VPC, and private cloud.

Building a Three-Layer AI Agent Matrix to Strengthen the Industry AI Agent Operating System

To address tasks at different enterprise levels, Baidu Intelligent Cloud has developed a comprehensive AI agent product matrix covering general-purpose, specialized, and customized agents.

The specialized agent matrix includes several capability products:

• Baidu FaMou: A self-evolving super AI agent
• Baidu MiaoDa: An application development AI agent
• Baidu ShengSuan: An ontology-building AI agent
• Hogee: A marketing AI agent
• Baidu YiJian: A visual AI agent
• Wenxin KuaiMa: An enterprise-level code AI agent
• Baidu YiJing: A digital human AI agent

Taking Baidu FaMou as an example, it follows a model of human-defined standards combined with continuous AI optimization, iterating to produce optimal solutions. In a collaboration with automotive company Arte, after 120 rounds of evolution, the prediction accuracy for vehicle drag coefficient reached 97.5%, improving simulation iteration efficiency by hundreds of times and reducing processing time from hours to seconds. FaMou has also demonstrated value in extreme engineering scenarios such as semiconductor chip layout optimization, solving complex multi-constraint engineering challenges.

Xin Zhou outlined four core requirements for enterprise customized AI agents:

1. Advanced Harness architecture
2. A robust and reusable skill ecosystem
3. Seamless integration with business systems
4. Enterprise-grade security and compliance assurance

Currently, customized AI agents have been deployed in large enterprises across multiple business sectors, with over 3,500 agents online and a daily token call volume of 20 billion.

To lower the development barrier for enterprises, Baidu has also launched the Baidu DaZi Development Platform, which supports agent configuration, operation, and full lifecycle management. The underlying technology is co-developed with Baidu DaZi, offering high concurrency, full-link observability, and flexible deployment options.

By forming an Industry AI Agent Operating System with general-purpose, specialized, and customized agents, and using Baidu DaZi as the super entry point, Baidu Intelligent Cloud enables downward scheduling of cloud resources and business tools, while supporting various business applications upward. The company is fully committed to helping enterprises improve operational efficiency and drive business innovation.

In closing, Xin Zhou expressed his hope to collaborate with all sectors of industry to continue exploring the practical application of AI in enterprises and unlock the full industrial value of artificial intelligence.