Earlier this 12 months, Google launched the Common Commerce Protocol (UCP), an open customary designed to allow AI brokers, companies, and fee suppliers to speak and transact by a shared framework. Then got here Common Cart, the corporateβs new clever buying hub that brings merchandise from throughout its ecosystem right into a single, AI-powered vacation spot.
In case you rewind a bit additional, Adobe reported that site visitors from generative AI instruments to U.S. retail websites jumped practically 1,300% YoY in the course of the 2024 vacation season, with sturdy development carrying into the 2025 season.
Throughout these turns of occasions, it might be tough to pinpoint precisely the place shopper conduct and platform innovation began to feed into one another. However as AI brokers, shopper platforms and fee suppliers race to attach with each other, retailers more and more discover themselves within the center, liable for managing transactions they’ve much less and fewer affect over.
The Architectural Dilemma
As a rule, integration is essentially being optimized on the stage of connectivity between methods, relatively than on the level the place selections are literally executed. AI platforms are steadily being enabled to speak with fee suppliers. However the best way a transaction is structured, routed, and managed throughout methods has not been basically reconsidered, additional pushing the issue onto the service provider. That is the place a key architectural query arises: the place ought to agentic fee logic dwell?
Whether it is embedded within the buying cart, the logic sits too near the platform layer, rising dependency and threat of lock-in. If it sits fully with fee service suppliers, retailers retain execution however lose the flexibility to form how transactions are routed throughout their very own methods.
What’s lacking in each approaches is a impartial layer that may interpret intent and coordinate execution with out being tied to both the interface the place the acquisition begins or the infrastructure the place it’s settled. In a single case, routing is constrained by platform logic, and within the different, itβs fragmented from the service providerβs broader commerce stack.
Putting agentic logic on the fee layer turns into the third possibility. Fairly than forcing retailers to decide on between platform dependency and payment-provider dependency, it creates some extent the place transaction selections stay beneath service provider management no matter how the acquisition is initiated or finally processed. Retailers can introduce agent-driven experiences with out redesigning their fee infrastructure, whereas persevering with to coordinate transactions throughout stock, order administration, and fee routing.
New Dangers Name for Clear Separation
Agentic commerce is outlined by larger autonomy for AI methods appearing on behalf of consumers. That autonomy, nonetheless, makes it more durable to differentiate AI-initiated transactions from customary fee flows, exposing core methods to new operational uncertainty.
As an illustration, an AI agent may full a purchase order after dynamically evaluating costs or making use of finances constraints set by the consumer, however the ensuing fee nonetheless enters the service providerβs system as a typical checkout occasion. With out separation, retailers lose visibility into how that transaction was initiated and whether or not it must be handled otherwise from a standard buy.
This reinforces the case for planning agentic fee logic on the fee layer, the place these flows might be remoted and managed independently. This implies figuring out AI-initiated transactions on the level of entry, permitting retailers to use completely different routing selections, fee suppliers, or controls, relying on the character of the transaction. Consider limiting purchases above, say, $100 or barring luxurious purchases if initiated by an agent. It additionally creates an area to check and adapt agent-driven purchases with out affecting the steadiness of the broader fee system.
Setting Operational Boundaries
The indistinguishability of agentic transactions mentioned earlier additionally creates a broader problem of classification inside present fraud methods. As a result of AI brokers intently resemble automated bots of their conduct, they’re usually handled as high-risk exercise by default in environments already saturated with AI-driven fraud makes an attempt, and fairly moderately so. Belief, subsequently, can’t be inferred from conduct alone. It must be explicitly signaled and acknowledged by every layer of the system.
This implies the buyer should clearly authorize the agent to behave on their behalf, and that permission should be recognizable not simply to the fee supplier but additionally to the service provider initiating the sale. On the identical time, the service provider should additionally have the ability to affirm that the agent is working beneath legitimate authorization, whereas additionally offering its personal verifiable identification to the agent.
In case you think about it this fashion, youβll discover that isolation alone just isn’t enough with out a corresponding mannequin for controlling entry to fee credentials. Since brokers require some type of delegated permission to transact on a consumerβs behalf, permission should be structured in a means that’s express, restricted, and verifiable throughout the transaction movement. Credential dealing with subsequently turns into a essential extension of payment-layer logic. In impact, fee particulars usually are not uncovered on to the agent however are as a substitute saved in safe environments and represented by tokenized credentials that act as managed proxies for underlying transactions.
These tokens outline the operational boundaries inside which agentic fee logic can operate. Fairly than granting open-ended entry to a fee technique, they will encode limits similar to spending thresholds, frequency, or validity home windows, guaranteeing that agent-driven transactions stay strictly inside predefined constraints. On this sense, what we are able to name βvaultingβ is the mechanism by which these constraints grow to be virtually enforceable.
Therefore, whereas the fee layer determines how transactions are recognized and ruled, tokenisation ensures that brokers can solely execute actions they’ve been explicitly approved to carry out.
Rising Amidst Rising Strain
Up till just a few years in the past, the concept of autonomous brokers making purchases in your behalf might have appeared like a distant chance. However right now, the U.S. B2C retail marketplace for agentic commerce is forecast to succeed in roughly $1 trillion in income by 2030. For retailers, the importance of that scale lies not within the quantity itself, however in what it means for visibility and management inside their very own methods.
The strain to maneuver shortly can price retailers the readability wanted to know what is definitely taking place in their very own methods. Of all the facility that granular visibility and management can vest upon retailers, essentially the most useful is the liberty and suppleness to begin small, be taught from early patterns, and scale consistent with confidence.
