At the Create2026 Baidu AI Developer Conference, Du Xiaoman Chief Technology Officer Zhang Wenbin noted that as building agents becomes easier, establishing a governance system for intelligent agents will become a core task, a challenge every team working on AI applications must address in the future.

Zhang Wenbin shared Du Xiaoman's practical experience in promoting large-scale agent applications in the financial sector and pointed out, "Over the past year, Du Xiaoman has deployed agents in multiple core business scenarios, achieving initial application results."

In customer service, Du Xiaoman's financial advisor agent provides 24-hour personalized service, increasing the one-time resolution rate for customer issues from 60% to 87% and boosting the credit usage rate by 15%. Previously, customers had to wait in line for human customer service; now, an agent that understands their needs is available at any time.

For risk control, the risk decision-making agent enables self-evolution of risk models and strategies, reducing the strategy iteration cycle from weeks or even months to hours and lowering credit delinquency risk by 12%.

In customer management, the customer insight agent accurately captures customer needs and timing, promptly meeting personalized requirements. 95% of its product optimization suggestions are adopted by employees, accelerating product improvement and innovation.

As agents are implemented in core business scenarios, Du Xiaoman has gradually developed a comprehensive financial-level agent infrastructure to support agent construction and operation. To enable every employee to use this financial-level agent infrastructure, Du Xiaoman built the one-stop AI application platform "Yuanli Platform"—not only providing complete capabilities for agent building but also integrating security, compliance, and permissions as a stable foundation. This allows each employee to easily build agents for various scenarios on the Yuanli Platform based on their professional expertise, transforming the agent infrastructure from supporting a few core agents into the AI capability foundation for the entire organization.

After the Yuanli Platform launched, it currently runs over 300 high-frequency intelligent agents, significantly improving work efficiency. For example, the anti-money laundering review agent, which previously relied on manual processes with low efficiency and was influenced by reviewers' personal experience, was built by a business team member using the Yuanli Platform's workflow orchestration and tool invocation capabilities in just one week. Since going live, it has reviewed over 20,000 cases with a direct adoption rate of 93%, reducing single-case review time to 50 seconds.