CAMBRIDGE, Mass. – At the 2026 Harvard College China Forum, North America’s longest-running student-led conference on China, Snaplii Founder and CEO Spencer Xu took the stage to make a pointed argument about the next era of commerce: as AI agents move from answering questions to completing tasks, the hardest problem they face is not intelligence; it is paying. Drawing on a decade building payment systems in North America, Xu detailed how Snaplii’s newly launched Agent-to-Merchant (A2M) Skill is built to solve that “last mile.”
“AI used to help people find information. Now agents do things for people,” Xu told the forum. “But the last mile of getting something done is payment, and that is exactly where today’s card rails break.”
From information flow to capital flow
Xu framed the global economy around three flows: information, logistics, and capital. Capital flow, he noted, has been his life’s work. He recalled speaking at Harvard three years earlier as the head of a Canadian company expanding into the United States, at a moment when every conversation was about AI. Back then, he admitted, he told the room that his company had “no AI and no interest in it,” only to discover the person seated beside him was the founder of a large language model. Since then, he said, his thinking has iterated sharply: the rise of AI agents has turned payment into the natural meeting point of all three flows.
“AI agents have no borders, and neither does information,” Xu said. “But payment has strong borders. Borders mean compliance, licenses, rules, banking systems, and real capital flow. So AI builders should do what they do best and own the information flow. Let us handle the licenses, the compliance, the chargebacks, and the money flow, and let’s learn together.”
Why agent payments break on legacy rails
Xu outlined why existing card networks were never designed for autonomous software. Bank authorization and risk-control systems, he explained, were built for human-paced, one-at-a-time purchases, not for agents that may fire many transactions per second. Chargeback liability is equally unresolved: if an agent completes a purchase and something goes wrong, it is unclear who is responsible, since the shopper did not click “buy” and the agent cannot be held liable. And because today’s agents still hallucinate and make mistakes, users are understandably reluctant to hand them real card credentials.
How Snaplii’s A2M Skill works
Snaplii’s Agent-to-Merchant Skill is designed to let AI agents complete authorized purchases while keeping a shopper’s bank and credit card details fully isolated from the agent. When an approved agent is ready to buy, Snaplii generates a one-time, transaction-level payment instrument drawn from the retailer’s electronic gift cards and the Prepaid Payment Network. Each credential is pre-funded, limited to a single authorized transaction, and cannot be reused, so there are no shared credentials and no persistent risk.
The A2M Skill is compatible with the Model Context Protocol (MCP), allowing it to be called directly from mainstream AI clients, including Claude and other MCP-compatible applications. On top of the security layer, shoppers can still earn up to 10% cashback on every transaction, stackable with existing deals and promotions.
“An agent should never touch your bank card,” Xu said. “With A2M, every purchase is pre-funded, isolated to a single transaction, and impossible to reuse. We take the merchant network and the regulated payment rails we’ve spent ten years building, and we make them safe for agents to use.”
A two-stage path to autonomous checkout
Xu described agent payments as unfolding in two stages. In today’s experimental phase, a human stays in the loop: the agent proposes a purchase, such as ordering a ride or a meal, and the person reviews and approves the payment before it goes through. The next stage, he said, is an agent a user trusts to decide independently what to pay for. That future, Xu argued, is only possible if the underlying payment layer tolerates no hallucination and cannot be blocked by legacy bank risk controls or chargeback disputes, the precise gap A2M is meant to close.
A decade of payments, now opened to agents
Snaplii is a digital wallet and smart-shopping app serving a large North American user base, with more than 350,000 users and over $100 million in annual transactions across 500-plus brands. Xu, a serial payments entrepreneur whose earlier venture SnapPay focused on merchant acquiring, positions Snaplii as a consumer wallet built on a simple promise of helping people save money, while giving merchants low-cost liquidity rather than just marketing traffic.
“Wallets and payments are usually a giants’ game, and we are small compared to the incumbents,” Xu said. “But new scenarios create new payment leaders. AI agents are that new scenario, and we intend to learn fast and build the payment layer they need.”
Xu invited founders building agentic commerce to collaborate with Snaplii: “If you’re building an AI agent that needs to pay in the real world, come talk to us. You do the new information flow; give us the licenses, the rules, the chargebacks, and the money flow.”
About Snaplii
Snaplii is a digital wallet and smart-shopping app that helps people save money while they shop across North America. The platform offers cashback rewards of up to 10% at more than 500 brands and serves over 350,000 users who transact more than $100 million annually. With its Agent-to-Merchant (A2M) Skill, Snaplii provides a secure, MCP-compatible payment layer that lets AI agents complete authorized purchases using one-time, pre-funded, non-reusable credentials, keeping shoppers’ bank and card details isolated from the agent. Learn more at snaplii.com.
Media ContactCompany Name: Fosmos consulting management co.,ltd.Contact Person: Matt.SunEmail: Send EmailCity: ShanghaiCountry: ChinaWebsite: https://mp.weixin.qq.com/s/6cDBikL9B7p3Jdvn-hOCBw
