Changes in how consumers shop and pay never happen all at once. The transformation doesn’t make a grand entrance. Change happens when the friction that consumers and merchants tolerate becomes harder to live with than the habit of sticking to the old way. That is when innovators step in, spot the opening and decide it is time to move forward. The signs of change always start quietly. Adoption curves begin to tilt upward. Expectations subtly shift. New behavior starts to settle in. Then ignition happens.
We have been here before, and the examples are familiar. Uber didn’t simply fix those “broken” taxi card readers. It replaced the entire mobility landscape with cleaner cars on demand and payments that disappeared into the background. Amazon didn’t just build a better checkout experience. It collapsed millions of products and a dozen checkout steps into a one-click-purchase marketplace that became the default starting point for eCommerce.
Both solved massive pain points. Both redefined values. Both forced the rest of the industry to decide whether to adapt or get left behind.
We are now approaching that same moment again.
This time, the shift is not about making payments invisible or shaving a few seconds off the checkout flow. It is about something much bigger: who — or what — makes the decision about what to buy and how to pay.
And where that journey begins.
Advertisement: Scroll to Continue
The End of Human-Only Decision Making
Agent-led commerce is often described as a revolution. It is actually the next logical step. Systems designed for human attention and human decision cycles are now interacting with autonomous agents that operate with far more context, speed, precision and consistency than people ever could. Researchers estimate that AI can process quintillions (hundreds of million trillions) of calculations per second. Humans, roughly 10 bits per second. Put simply, it would take hundreds of people working nonstop to match what a single AI model can analyze in under a minute.
Naturally, when agents become part of the buying decision, the old interfaces break down. Screens, forms and flows built for people introduce friction that agents neither need nor use efficiently. That mismatch becomes the next competitive battleground.
Two models are emerging, and both point in very different directions.
|
The question is no longer whether agents will transact. It is how they will do it. And who owns the data that informs those decisions and the liability when something goes haywire. And, importantly, who captures the economics when an agent completes a purchase.
Whoever solves it will shape the next era of commerce.
Two models are emerging, and both point in very different directions.
OpenAI and Stripe launched the Agentic Commerce Protocol (ACP) in September 2025. It puts merchants and their existing infrastructure at the center. That same month, Google introduced the Agent Payment Protocol (AP2), which positions discovery and product data as the control point.
These two approaches reflect fundamentally different ideas about where value and control should sit in agentic commerce. ACP is a checkout-centric model that routes agents into the merchant’s existing infrastructure. AP2 is a mandate- and knowledge-graph-centric model that defines how agents express intent and authorization across ecosystems.
To understand what could scale and what may stall, you have to look at what each model demands from the ecosystem to get commerce run by agents off the ground.
The Infrastructure That Had to Be Built First
Before any commerce model could flourish, agents needed something foundational: a governance and control layer. They needed a safe way to take actions across systems without breaking workflows or tripping security boundaries. That foundation is the Model Context Protocol (MCP), launched by Anthropic in November 2024 as a neutral layer that lets agents operate inside enterprise systems securely and at scale.
MCP is simply the plumbing that allows ACP and AP2 to make agentic commerce a reality.
|
Banks, acquirers, processors and networks have already begun using MCP for identity verification, chargeback triage, settlement, merchant onboarding and fraud. These early initiatives signal that enterprises are preparing agents to run real workflows inside their systems. The analogy is simple. HTTP made reliable communication possible across the early internet. MCP standardizes safe and structured actions for agents.
MCP is not trying to control the demand funnel or the transaction economics. It is simply the plumbing that allows ACP and AP2 to make agentic commerce a reality. While ACP and AP2 compete for influence over transaction flows and merchant relationships, MCP ensures that agents can interact with banks, processors and merchant back ends without introducing new operational or security risks.
Only once MCP established this baseline could real competition begin.
Model One: Merchant-Centric Agentic Commerce Protocol (ACP)
The Agentic Commerce Protocol emerged from the partnership between OpenAI and Stripe. It brings together the dominant conversational interface and one of the most widely adopted payment platforms. Together they built a protocol that lets agents complete purchases inside the AI experience while keeping the merchant’s current tech and payment stack intact.
Stripe brings decades of experience in digital payments, compliance and settlement workflows. OpenAI brings deep understanding of how conversational intent becomes transaction ready. Their bet is straightforward. ACP gives agents a standardized way to send orders into systems that already work for human checkout. Agents discover products, build carts, authorize payments and trigger fulfillment through flows merchants already trust.
ACP gives agents a standardized way to send orders into systems that already work for human checkout.
|
This low coordination burden is why ACP is drawing partners like Instacart and DoorDash. Stripe aggregates millions of merchants. OpenAI aggregates hundreds of millions of users. ACP bridges the two without forcing merchants to rebuild catalog structures or operational logic. Other processors are now treating ACP as the template for “agent-ready commerce.”
But ACP’s advantages come with new challenges.
Delegated payments, a key part of ACP, are fully implemented only by Stripe. Other PSPs need months to reach parity.
In the meantime, many processors and networks are launching ACP-style variants with the same agentic thesis: tokenized credentials running over existing card and account-to-account rails, with disputes resolved inside current network rules. That means that merchants still have to implement ACP or a variant, or more than one, stitching multiple flavors of agentic commerce into their stack. And integrating with and supporting them.
Model Two: Data Centric Agentic Payments Protocol (AP2)
Google’s Agent Payments Protocol starts somewhere entirely different. It is a mandate and knowledge-graph framework, not an end-to-end transactional rail. Google has spent decades building a global product graph through Shopping, Merchant Center and schema.org. AP2 formalizes that graph and asks merchants, PSPs and platforms to feed it structured data and adopt a common language for agent mandates. It centers on cryptographically signed purchase mandates, verifiable credentials and a universal product index. Payment rails are treated as interchangeable plumbing.
AP2 launched with 60 partners across eCommerce, payments, crypto and enterprise software. Its goal is to be a shared protocol for how agents express intent and authorization. Crypto platforms see AP2 as a quiet path to participate at scale. If agents can settle over crypto rails behind the scenes, merchants never need to support those payment methods directly.
AP2 asks merchants, PSPs and platforms to feed it structured data and adopt a common language for agent mandates.
|
But this design creates a much higher coordination burden.
AP2 expects merchants to maintain a structured catalog that may not match their internal schemas. For merchants with large SKU counts, this becomes significant operational work.
And AP2 cannot guarantee that agents will query its index or that merchants will see incremental demand. Agents decide whether to use AP2. Merchants cannot treat it as a steady demand source.
How Merchants Will Decide
From the merchant’s perspective, ACP and AP2 represent very different tradeoffs.
In theory, ACP fits cleanly into today’s checkout and order-management environment. Merchants keep their product data, processors and fulfillment logic. Agentic becomes simply another way that orders enter systems they already trust. The primary effort lies in implementing the protocol(s) and managing delegated payments, not rebuilding catalogs.
AP2 prioritizes global consistency and agent portability across transacting environments. It emphasizes mandates, discovery through a universal product index and cryptographically verifiable intent. It needs merchants to align their data to this structure without promising demand in return. AP2 asks merchants to build and maintain a new structured product layer and ongoing data hygiene consistent with their knowledge graph. Conversion happens through processers that have adopted the AP2 protocol.
AP2 also inherits Google Shopping’s persistent limitation: strong discovery, weak transaction confidence. And a business model that is built to optimize advertising, not commerce.
AP2’s Google Shopping Hangover — and Headache
Google Shopping has existed since 2002. It has one of the largest product listings in the world and appears in billions of search queries. Yet it never became a marketplace. It never built the trust layer, fulfillment infrastructure, dispute processes or vetted seller network needed for end-to-end commerce. Google Shopping remains what it is intended to be: a lucrative, ad-driven discovery engine that directs traffic to the merchant, who will hopefully convert that visit to a sale.
Merchants exposed some of their inventory to Google Shopping because it drove human clicks, often supported by search ads. But agent-led commerce removes the ad-funded funnel entirely. If AP2 can’t show that it drives completed transactions, merchants may not be enthusiastic about investing in another product feed. Especially one that is disconnected from their existing order and payments systems. And one where the transacting economics remains TBD.
If AP2 can’t show that it drives completed transactions, merchants may not be enthusiastic about investing in another product feed.
|
To make AP2 work, Google would need to build (or rely on partners to build) something it has historically avoided: the hard, unglamorous, low-margin infrastructure that takes a product listing and turns it into a completed purchase, inside a Google Marketplace.
As all of this unfolds, ACP will continue to evolve. Not every platform will update on the same timeline. Merchants will find themselves supporting multiple protocol versions and PSP-specific variants whether they planned to or not.
Meanwhile, the number of agents seeking access to merchant systems will keep growing. Each one must be vetted, contracted and cleared from a security perspective. An abstraction layer that smooths out these differences and sets consistent standards for agent behavior may give merchants a way to participate without getting buried in agent fragmentation.
Igniting Agentic Commerce
Commerce innovations have never ignited quickly. Card networks, eCommerce, mobile wallets and one-click checkout all took years to evolve from something new into everyday behavior, even with familiar rails and payment methods to anchor the “new.”
Agentic commerce will follow that same pattern. The difference now is that the rails, the interface and the decision-maker are all changing at the same time.
ACP and AP2 face very different ignition paths. ACP rides on existing merchant systems, processors, issuers and risk frameworks, so the jump from pilot to scale is, at least in theory, short. Agents route orders into infrastructure merchants already trust, and delegated payments settle inside compliance flows everyone understands.
AP2 has a tougher lift. It requires the ecosystem to adopt a new discovery and representation layer, restructure catalogs around Google’s schema and rely on agents voluntarily querying Google’s product graph. At the same time, Google must build transactional and dispute-handling capabilities it has never owned. The more coordination required, and the less guaranteed the demand, the tougher the slog to ignition becomes.
Then there is the quiet elephant in the room: liability.
|
Igniting agent-led commerce is not just a technical milestone; it is an economic one. Agents don’t simply transact, they choose. And the path they choose depends on who they trust, what incentives shape those choices and how the ecosystems behind them intend to make money. Every major commerce shift has been anchored by a shift in trust: card networks scaled when liability became predictable; Amazon became the default marketplace when it proved more reliable than individual sellers; and Uber took off when riders trusted an algorithm to deliver a more reliable outcome than a dispatcher. Agentic commerce will follow the same pattern, but with a twist. Here, trust extends beyond safety and liability into something deeper. The business model that determines how discovery is surfaced and prioritized.
Search results and product recommendations have always reflected monetization models, but agents heighten the stakes. They compress discovery, evaluation and checkout into a single autonomous action. If an agent’s choice is influenced by advertising, sponsorship or retailer-owned media networks, then trust may be called into question. Google has refined the advertising-driven discovery model for decades, and AP2 naturally sits atop that legacy. Walmart is rapidly expanding its retail media capabilities, and agents could easily become the next frontier of monetizable demand shaping. OpenAI and Stripe, on the other hand, designed ACP around transaction flows rather than discovery economics, which, at least for now, limits the concern that paid placement might distort what an agent recommends. That could certainly change.
This matters because agents operate at machine scale. A small economic nudge, when applied across millions of autonomously completed transactions, can shift market share overnight. For agentic commerce to ignite, ecosystem participants must trust that agent recommendations reflect user intent, not necessarily platform incentives. When an agent surfaces a product because it has been paid to do so, but presents the action as optimal, consumer trust can erode. When merchants suspect that discovery is shaped by advertising fees rather than relevance, their appetite to invest in new protocols weakens. When issuers or networks cannot understand why an agent made a choice, their ability to manage risk and liability diminishes. Ignition stalls whenever the incentives of the system diverge from the expectations of the ecosystem.
In the end, the real competitive differentiator may not be technical at all. It may be the model that convinces the ecosystem that trust, incentives and outcomes are aligned. Business models will ultimately determine whether agent-led commerce reaches ignition. And until the economics of discovery become as transparent and aligned as the mechanics of the transaction itself, any model that cannot demonstrate that alignment will find ignition, and long-term success, are harder to achieve.
Until NEXT time.
Join the 19,000 subscribers who’ve already said yes to what’s NEXT.
PYMNTS CEO Karen Webster is one of the world’s leading experts in payments innovation and the digital economy, advising multinational companies and sitting on boards of emerging AI, healthtech and real-time payments firms, including a non-executive director on the Sezzle board, a publicly traded BNPL provider.
She founded PYMNTS.com in 2009, a top media platform covering innovation in payments, commerce and the digital economy. Webster is also the author of the NEXT newsletter and a co-founder of Market Platform Dynamics, specializing in driving and monetizing innovation across industries.
