According to PYMNTS.com, the race to control AI-driven commerce is crystallizing around two competing protocols launched in September 2025. OpenAI and Stripe introduced the Agentic Commerce Protocol (ACP), a model designed to plug AI agents directly into a merchant’s existing checkout and payment stack. That same month, Google launched the Agent Payment Protocol (AP2), a framework centered on its global product knowledge graph and cryptographically signed purchase mandates. These approaches reflect a fundamental split over whether value in agentic commerce sits at the point of checkout or the point of discovery. The entire competition is built on a foundational layer called the Model Context Protocol (MCP), launched by Anthropic in November 2024, which standardizes how agents safely interact with enterprise systems. Early partners like Instacart and DoorDash are already aligning with ACP, while AP2 launched with 60 partners across eCommerce and crypto.
The Plumbing War Before The Money War
Here’s the thing that’s easy to miss: none of this agentic commerce stuff happens without boring, secure plumbing. That’s what Anthropic’s Model Context Protocol (MCP) is. Think of it like HTTP for AI agents—a neutral layer that lets an AI safely log into a bank’s system, verify an identity, or trigger a settlement without causing a security nightmare. It’s not sexy, but it’s absolutely critical. And because companies like banks and processors are already using it for real back-office work, it means the stage is actually set. MCP solved the “how do we let agents in the door” problem. Now, ACP and AP2 are fighting over what the agents do once they’re inside, and who gets to guide them.
Two Philosophies, One Big Choice
So, what are merchants actually choosing between? It’s a classic tech dilemma: evolution versus reinvention. The OpenAI/Stripe ACP model is basically evolution. It says, “Hey merchant, your Stripe-powered checkout works great for humans. Let’s just teach AI agents to use that same pipe.” The agent does its shopping, then sends a standardized order into your existing order management system. You keep your catalog, your logic, your everything. The friction is low, and that’s why it’s getting traction.
But Google’s AP2 is a full reinvention play. It starts with discovery, not checkout. Google wants every product in the world described in a structured way inside its knowledge graph, and it wants agents to use cryptographically signed “mandates” to shop across it. The payment part is treated as interchangeable backend plumbing. Sounds powerful, right? But look at the ask. It requires merchants to rebuild and maintain a complex product feed for Google—again—with no promise of actual sales. It inherits the same old Google Shopping problem: amazing at showing you stuff, historically bad at actually closing the transaction and handling the messy realities of commerce.
The Google Problem
And this is Google’s core challenge. Google Shopping has been around since 2002. It’s massive for discovery. But it never became Amazon. It never built the trust, fulfillment, and dispute resolution guts of a real marketplace because that’s low-margin, hard work. Its business is ads. So AP2 is, at its heart, an attempt to extend that ad-driven discovery engine into the AI age, but without the ads. The value proposition for merchants is suddenly murky. If an AI agent uses AP2 to find my product but then completes the purchase elsewhere via ACP, what did I get from maintaining that AP2 feed? Not much. Google would need to build the commerce infrastructure it’s always avoided to make AP2 truly compelling, and that’s a huge bet.
What This Means For Everyone Else
For developers and enterprises, this is a new integration headache. You’ll likely need to support multiple “agent-ready” protocols, just like you had to support multiple browsers or payment methods. For payment processors and crypto platforms, it’s a land grab to become the default rail inside these new flows. And for users? We probably won’t notice the protocol fight at all. We’ll just see AI assistants that get better at completing complex tasks, like re-stocking a pantry or planning a trip, without constant hand-holding. The winner of this protocol struggle won’t be the one with the smartest AI, but the one that builds the most indispensable economic layer. Will it be the checkout, or the discovery engine? The whole next era of digital commerce hinges on that answer.
