Roughly 23% of US consumers have already made a purchase with the help of an AI assistant in the past month. That figure, from Morgan Stanley, is the part of the agentic commerce shift that has already happened. The part that hasn’t is the work merchants need to do on the other side of those transactions: making their catalogs discoverable inside AI environments, routing the resulting payments intelligently, and getting visibility into what agents are actually doing once they arrive.
For most enterprise merchants, the checkout that handles these flows was designed for a customer tapping through a mobile cart. The gap between what those checkouts can do today and what an AI-driven purchase actually requires is widening every quarter.
This article looks at what agentic commerce means for merchants, why most existing payment stacks are not equipped to handle it, and what a practical path forward looks like for businesses that want to be discoverable inside AI environments without rebuilding their stack from scratch.
What agentic commerce actually is
Agentic commerce describes a category of transactions where an intelligent system, rather than the customer’s own clicks, drives the discovery, evaluation, and execution of a purchase. The customer still authorizes the spend, but the path from intent to checkout runs through an AI assistant, a chat interface, or an autonomous agent acting on the customer’s behalf.
The earliest forms are already in market. Consumers ask ChatGPT for product recommendations and complete the transaction inside the conversation. Voice assistants reorder household goods on a schedule. Specialized agents compare flights, book hotels, and process payments without the customer ever loading a merchant’s website. Each of these flows starts somewhere the merchant does not control and ends with a payment that the merchant’s existing stack was never designed to handle.
The strategic implication is direct. Merchants who do not appear inside the AI environments will not be considered by the agents acting on customers’ behalf. Being unsearchable in the AI layer is becoming the new equivalent of being unsearchable on Google.
Why most payment stacks are not ready
The payment stack that most enterprise merchants use today was built for a different shopping pattern. A customer arrives on a website, adds items to a cart, enters card details, and clicks pay. The merchant’s PSP authenticates the transaction, sends it to the acquirer, and the funds eventually settle.
Agentic transactions break several assumptions in that flow at once.
The first is that the customer is present at checkout. An agent may complete a purchase on behalf of a customer who is not actively looking at the screen, which complicates strong customer authentication, 3D Secure prompts, and the entire concept of checkout abandonment.
The second is that the merchant controls the storefront. When the storefront is a ChatGPT conversation or a voice assistant’s response, the merchant does not own the surface where the customer is deciding. The product catalog, pricing logic, and payment options all need to be discoverable and executable from outside the merchant’s domain.
The third is that authentication, tokenization, and routing happen inside a single relationship with a single payment provider. Agent-driven transactions involve handoffs between AI platforms, the merchant’s stack, and the payment processor, with credentials and tokens needing to remain secure and valid across the entire chain.
The fourth, and the one most often overlooked, is that the merchant has visibility into how the transaction reached them. Without specific tagging and analytics, an agent-driven transaction looks identical to a regular checkout in the merchant’s data, making it difficult to measure performance, understand customer behavior, or optimize the experience. This is where a payment orchestration platform (POP) becomes valuable. By providing a centralized layer across payment providers, methods, and channels, a POP such as Gr4vy can help merchants identify, track, and analyze agent-driven transactions separately, giving them the visibility needed to make informed decisions as these new commerce models emerge.
What merchants actually need to participate in agentic commerce
The capabilities that make agentic commerce workable do not live in one product today, which is why most merchants have a gap. Discoverability has to come first: the product catalog has to be accessible to AI agents in a structured, machine-readable format, with current pricing, availability, and purchase logic. This sits closer to API publishing than to traditional SEO, and the standards are still being defined by the platforms themselves.
Once an agent initiates a transaction, the payment has to flow through whichever PSP or acquirer is most likely to approve it for that specific customer profile, geography, and payment method. Routing rules, dynamic retries, fraud screening, and provider failover all need to work the same way they do for a customer-initiated transaction. The transaction also has to clear a security and compliance bar that scales to agent-initiated flows, which means card data tokenized in a way that works across providers and across the merchant-to-agent boundary, inside a PCI DSS Level 1 certified environment, with authentication that handles both customer-present and agent-initiated cases without falling back to the merchant’s engineering team to solve case by case.
The capability that ties the other three together is visibility. Merchants need to be able to tag agent-driven transactions separately, monitor their authorization rates, identify where they are losing conversion, and adjust routing based on real performance data. Treating agentic transactions as a black box is how merchants lose money quietly for quarters at a time.
The orchestration layer as the practical path forward
Building all of this in-house is technically possible, but the engineering effort is substantial. A functioning agentic commerce setup requires API publishing to multiple AI platforms, multi-PSP routing, a provider-agnostic Vault, Network Tokenization management, agent-aware authentication, fraud orchestration, and reporting that distinguishes agent flows from human ones. Each of these is its own discipline, and the standards governing them are still moving.
This is where payment orchestration becomes the natural integration point. An orchestration platform sits above a merchant’s existing PSPs, processors, and fraud providers, and gives them a single layer to add new transaction types without replacing what already works. For agentic commerce specifically, this means a merchant can keep their existing acquiring relationships, their current fraud rules, and their established checkout, while adding the agent-initiated flow as another channel that the orchestration layer routes intelligently.
Gr4vy, the cloud-native payment orchestration platform, recently released a structured kit aimed at this exact problem. The Agentic Development Kit (ADK) provides merchants with the infrastructure and step-by-step guidance to enable their storefront inside AI environments like ChatGPT, orchestrate the resulting transactions across more than 400 PSPs and payment methods, and maintain control over routing, security, and performance. The ADK is designed to work without replatforming and without requiring the merchant to drop their existing payment providers.
Three things stand out about this approach for businesses evaluating their options. The first is that it preserves what already works: the merchant does not have to swap out their PSPs or rebuild their checkout, because the orchestration layer routes agent-driven transactions to the same processors the merchant already uses, with the same fraud rules and the same compliance posture. The second is that it gives the merchant a separate lens on agent traffic. Transactions initiated by an agent are tagged and trackable, with authorization rates and conversion measured independently from human-initiated flows, which is what makes ongoing optimization possible rather than guesswork. The third is that it runs inside a PCI DSS Level 1 certified environment with full authentication, which removes the compliance burden that would otherwise sit on the merchant’s engineering team. For enterprises, this is often the deciding factor in whether agentic commerce projects make it past legal review.
Three criteria that separate workable solutions from new lock-in risks
For merchants assessing their options, a small number of criteria separate workable solutions from ones that will create new problems.
The first is that the solution has to be provider-agnostic. Any approach that locks the merchant into a single PSP or a single acquirer for agent-driven transactions is recreating the vendor lock-in problem that orchestration was designed to solve in the first place. The second is that it has to preserve the merchant’s existing stack: replatforming for agentic commerce is a signal that the vendor’s architecture is not designed for the multi-channel reality most enterprises operate in. The third is real visibility rather than just routing. Tagging, analytics, and the ability to A/B test routing rules across agent flows are what make the difference between a project that pays back in the first quarter and one that becomes a maintenance burden.
There is also a question of timing that sits underneath all three. Agentic commerce protocols are still being defined by Visa, Mastercard, OpenAI, Anthropic, and others, and the infrastructure layer has to be flexible enough to absorb those changes without forcing the merchant to rebuild. At the same time, the pace at which AI shopping behaviors are emerging means that a six-month implementation window is functionally a missed market. Solutions that can be live in weeks, with clear documentation, are the ones that match the actual speed of change.
The takeaway
The shift to agentic commerce is closer than the planning horizons of most e-commerce roadmaps allow for. Consumers are already buying through AI assistants. Platforms are already publishing the protocols. The merchants who treat this as a 2027 problem are likely to discover in 2026 that their competitors are already discoverable inside the AI environments their shared customers are using.
The good news for businesses with established payment stacks is that participating in agentic commerce does not require starting over. The orchestration layer that solves multi-PSP routing, cross-border optimization, and authorization rate uplift for traditional ecommerce is the same layer that can absorb agent-initiated transactions, provided it is designed to do so.
For merchants beginning to plan their approach, the practical first step is auditing the current stack against the four capabilities above: discoverability, orchestration, security, and visibility. The gaps that audit reveals will define whether agentic commerce becomes an opportunity to capture new revenue or a category the business gets locked out of by the time the standards stabilize.
Gr4vy‘s no-code payment orchestration platform is the only one that empowers enterprises with full control to automate, customize, and optimize their payment strategy effortlessly, while eliminating the risk of a single point of failure. Discover how the Agentic Development Kit can turn AI-driven commerce into a measurable channel inside your existing stack.
Related Categories