AI Agent Completes First B2B Payment in Greater China Trial

LianLian, a payment platform, and Visa, the payments network, conducted the first live B2B agentic transaction in Greater China, marking a shift in how artificial intelligence is deployed in business operations. The transaction centered on LianLian's LoopXPay agent, which operated with support from Visa's agentic commerce solutions.

In the pilot workflow, the agent performed a complete supplier payment cycle without human intervention at each step. It began by mapping the purchasing requirement, then suggested and compared potential suppliers, placed an order, and executed the payment—all within a unified system and operating under defined spending and approval parameters. The seamless integration of these steps illustrates how agentic AI differs from previous AI deployments: rather than assisting humans in making decisions, the agent made and executed decisions autonomously.

The motivation for the trial centers on the pain points that small and medium-sized businesses face in managing accounts payable and accounts receivable. These organizations typically operate with limited staff and must devote significant human resources to routine tasks: managing credit lines, reviewing invoices, reconciling payments, and resolving exceptions. By automating these tasks, the pilot aimed to show how business owners could redirect their attention to growth and strategic work. According to the article, this potential has driven substantial business investment in both generative AI and agentic AI systems.

Moving such systems from controlled trials into production environments, however, presents substantial challenges. Agentic transactions executed within critical business workflows require robust governance and security infrastructure—controls that enable oversight while still allowing organizations to deploy and refine systems iteratively. A particular concern is fraud: accounts payable and accounts receivable processes have long been targets for cybercriminals, and as AI agents assume operational authority, criminals are expected to adapt their tactics to target the agents themselves rather than manipulating human users. Additionally, the industry has identified a structural gap: AI agents currently lack a native, secure mechanism to transact and interact with one another in systems originally designed for human participants. Despite these open questions, the article suggests that as more trials are conducted globally, the agentic commerce ecosystem will mature and help define how AI-driven transactions operate at scale.

The pilot represents a significant step in the evolution of AI from a tool for human productivity into an autonomous operator within critical business workflows. Unlike previous applications of generative AI—which have primarily assisted humans or augmented decision-making—agentic AI executes end-to-end transactions within defined parameters. For accounts payable and accounts receivable functions, which have historically consumed significant administrative resources in small and medium-sized enterprises, this shift could reshape operational efficiency.

The trial also exposes the tension between automation's potential and the operational reality of deploying such systems at scale. The article notes that AR/AP workflows have become targets for fraud partly because of AI-driven schemes developed by cybercriminals. As AI agents take on autonomous transaction authority, the attack surface expands: criminals will no longer need to manipulate human judgment but can instead develop tactics designed to compromise the agents themselves. This suggests that governance and security infrastructure—oversight controls, audit trails, and exception handling—will become as critical to agentic commerce as the AI models themselves.

No comments yet. Be the first to share your thoughts!