How well do you know your agent?

For years, the payments industry rested on a simple comfort: a human thumbprint on every purchase. A tap of a card. A biometric scan. A 2FA prompt. That pattern held through contactless, digital wallets, and one-click checkout. Agentic commerce breaks it.

The human is no longer at the glass, authorizing in the moment. Authority is delegated in advance to an AI agent that can discover, decide, and attempt to transact on a consumer's behalf.

The question is no longer whether a payment can be processed. It is who, or what, is authorizing it, what that actor is allowed to do, and who is liable when something goes wrong.

That gap has a name: Know Your Agent, or KYA. And it points to a larger shift. The control point of commerce is moving away from the checkout page and toward the layer that can verify an agent, enforce what it is allowed to do, and route its transaction through the right rails. Whoever operates that layer operates agentic commerce.

Agentic commerce is already here

In 2026, agentic commerce is no longer a thought experiment confined to conference panels. AI sits inside the shopping journeys of hundreds of millions of consumers, who use general-purpose models and retailer assistants to search, compare, and narrow purchase decisions. In some cases they already let AI help at checkout.

Consumers have invited AI into the decision layer and are hesitating at the transaction layer. Nuvei's How Agents Pay research shows that 58% of consumers find AI useful for researching products and comparing prices, and 39% say they are using AI while shopping more often than they were only three months ago. Yet only 27% would feel comfortable storing payment credentials on an AI platform, and 28% would feel comfortable letting it complete a purchase automatically.

The buyer of record at checkout is now, more often, an agent acting on behalf of a person, and adoption is moving faster than the infrastructure built to govern it. McKinsey estimates that agentic commerce could drive trillions of dollars in global retail spend by 2030, which turns today's gaps in trust, liability, and standards into an urgent problem rather than a theoretical one.

Knowing your customer is no longer enough

Traditional payment systems were built for a human buyer. A person sees the transaction, recognizes it, authorizes it, and can dispute it later. Fraud controls, dispute rules, and liability frameworks were all designed around that pattern. When the buyer is an AI agent, the pattern breaks.

The control point in commerce is shifting. It is no longer the checkout page. It is the layer that can interpret an agent's intent, verify its mandate, and route the transaction through the right rails with the right controls.

That layer has to answer a different set of questions.

• Is this agent what it claims to be?
• On whose behalf is it acting?
• What has the consumer authorized, in amounts, categories, time windows, and named merchants?
• And what evidence survives a dispute when no human ever pressed pay?

Know Your Customer has been foundational in financial services for decades. It verifies the human behind the account. In an agentic world that work remains necessary, and it is no longer sufficient. The system now needs a way to verify the agent itself, its identity, its mandate, and its behavior, before it touches the payment flow. That is what KYA is for. It extends the compliance logic of KYC from humans to the non-human actors entering commerce, and it establishes an agent's identity, authority, and behavioral boundaries before the agent can initiate or complete a transaction.

Early work is underway. New protocols are starting to carry trust signals about agents, including their provenance and capabilities. The market is still defining what production-grade agent identity, mandate handling, and reputation scoring should look like, and how that information should move between AI platforms, merchants, payment providers, and issuers. Until KYA is built into the infrastructure, agentic commerce stays promising and incomplete.

When standards multiply

The identity problem would be hard enough on its own. The protocol landscape is multiplying at the same time. Several agent frameworks are emerging to define how AI agents talk to merchants, payment providers, and one another in order to discover, negotiate, and complete transactions. Each speaks a slightly different dialect. From a merchant or PSP's point of view, the picture is familiar and uncomfortable: no one wants to rebuild a payment flow five times to keep pace with the latest protocol.

What the market lacks is an execution layer that sits between agents and payment rails. One that abstracts the protocol differences, enforces mandates and limits, and translates an agent's intent into trusted, local transactions. That layer cannot be a cosmetic patch or a single line on a roadmap. It has to be operational, certified, and connected to the places where commerce already happens: gateways, acquirers, local payment methods, and issuer rails.

This is where the control point sits: at the execution layer that brings agent identity, mandate enforcement, and local routing into one place. It belongs at the PSP, gateway, and acquirer level, because that is where trust signals and money movement already meet, below the model and beneath the checkout page.

Local still wins, even for agents

The medium is new. Customer preference is not. A consumer in São Paulo who asks an AI assistant to reorder groceries still expects to pay with Pix. A voice-driven booking in Mumbai still needs UPI. A Perplexity-assisted purchase in the United States still has to work with the method the buyer prefers. Agentic commerce is global by design. Commerce stays stubbornly local at the point of payment.

If the acquiring path is wrong, the preferred method is missing, or local risk rules are ignored, the transaction fails as fast as it would for a human shopper. In agentic commerce that failure costs more, because the promise of the model is speed and delegation. When the agent cannot complete the purchase on the right rail, it does not wait. It moves to the merchant that can.

这就是为什么信任与本地执行密不可分。它们其实是同一个问题从两个不同角度的体现。没有执行能力的信任注定会失败，因为一个经过验证但无法在正确通道上付款的代理，最终仍会失去这笔交易。而没有信任的执行则充满风险，因为在没有经过验证的授权情况下完成的交易，无异于一笔随时可能爆发的负债。代理式商业要求这两者必须在同一位置、由同一层来解决。

从接受到信任

在上一代支付时代，核心挑战在于“受理”：无论商户身处何地，都能接收、转接并结算来自任何卡片或支付方式的款项。而在代理时代，核心挑战则是“信任”。如果代理购买了错误的商品，责任该由谁承担？当代理重复进行同一笔交易时，会发生什么情况？当一名代理为整个团队预订差旅时，如何执行每位用户的限额？当消费者根本没有点击确认按钮时，所谓“友好欺诈”又会表现为何种形式？

目前，能给出完整答案的服务商寥寥无几。连接到代理商其实是容易的部分。真正具备优势的是那些能够在一处完成信任验证、确保交易意图并利用正确的本地支付通道完成交易的服务商。 像Nuvei 这样的服务商，已在 50 多个国家开展本地收单业务，覆盖 200 多个国家和地区，支持 720 多种支付方式，其出发点正是代理商在决策另一端所需的支付通道。将这种覆盖范围与代理商身份验证及授权执行相结合，商业活动的核心问题便发生了转变——从“我们能否接受这笔支付”转变为“我们能否信任这位代理商，基于该授权，通过这些支付通道完成交易”。

数字商业的第一阶段侧重于接受度。下一阶段则侧重于委托信任。最终胜出的基础设施不仅能转移资金，还能决定何时转移资金，以及依据谁的授权进行转移。

‍