Ant Group researchers: AI agents push payments from click-to-pay to decide-to-pay
Ant Group Research Institute's Li Zhenhua and Feng Jiaqi argue in Peking University Finance Review that the agent economy is shifting payment competition from payment digitalization, which serves people, to transaction intelligence, which serves agents, and that China and the US are starting from the same line on this new dimension. Drawing on the IMF's three-layer architecture, they frame the core challenge as reconciling probabilistic, adaptive model reasoning with deterministic authorization and settlement, while noting that most global agentic payment protocols such as UCP, ACP, TAP and MPP are still in the specification stage with no products at scale. The article also frames risk governance as "1+3": existing payment regulatory frameworks plus layered management, dynamic continuous oversight, and technology-driven controls, mapped to HITL, HOTL and HOOTL human-agent collaboration modes.
The piece maps how agent-initiated payments are reshaping global payment competition and protocol standardization, with payment service providers such as Stripe positioned as the control-and-authorization layer between models and clearing networks.

随着智能体经济的到来,AI正从“能说”演进到“能做”,从信息交互演进到代理行动。过去两年,大模型展示了“能说”能力,但用户期待已不限于获取信息,而是希望AI直接完成任务、解决实际问题。2025年被业界视为AI应用元年,全球AI应用成熟度评价从关注模型参数规模转向关注“执行成功率、合规可控性、端到端闭环能力”,......
本文来自微信公众号: 北大汇丰PFR ,作者:李振华 冯佳琦
随着智能体经济的到来,AI正从“能说”演进到“能做”,从信息交互演进到代理行动。过去两年,大模型展示了“能说”能力,但用户期待已不限于获取信息,而是希望AI直接完成任务、解决实际问题。2025年被业界视为AI应用元年,全球AI应用成熟度评价从关注模型参数规模转向关注“执行成功率、合规可控性、端到端闭环能力”,IDC预测到2030年45%的组织将大规模编排自主智能体。
#Ant Group#Stripe#Google#OpenAI#Visa#Mastercard#Alipay#China UnionPay#JD Digits#International Monetary Fund#Amazon Rufus Auto Buy#IMDA