Agentic commerce stopped being a roadmap slide in April 2026, when OpenAI and Stripe shipped the Agentic Commerce Protocol and Google's Agent Payments Protocol moved under FIDO Alliance governance. Stripe, Adyen, and Checkout.com now support both networks. ChatGPT, Copilot, and Gemini can already complete a purchase without the shopper ever leaving the chat window. Most ecommerce and B2B checkout flows have no idea an agent is standing at the register.
This is not a future problem you can staff up for next quarter. Agents are evaluating your product pages, reading your feeds, and attempting checkout right now. If your catalog is not structured for machine consumption and your payment stack has no agent-facing hook, you are invisible to a channel that is already converting for your competitors.
The Protocol Stack You Need to Stop Confusing
Three acronyms get thrown around interchangeably, and that confusion is costing teams months of implementation time.
ACP is the commerce layer. Built by OpenAI and Stripe, it lets an AI agent execute a full checkout inside the chat surface itself, using Stripe's Shared Payment Token so the agent never touches raw card data. The transaction happens entirely inside ChatGPT or whatever ACP-compatible interface the shopper is using.
AP2 is the consent layer. Originated by Google and now governed by FIDO Alliance, it proves who actually authorized a purchase through cryptographically signed mandates, so a merchant can verify the human behind the agent approved this specific transaction at this specific price.
MCP is the discovery layer. The April 2026 ACP spec added an integration hook for Anthropic's Model Context Protocol, so an ACP-aware agent can discover that your store supports agentic checkout the same way it discovers any other tool.
These are not competing standards fighting for a winner. A real agentic purchase uses all three: MCP to discover you exist, ACP to execute the transaction, AP2 to prove the human meant it. Teams that pick one and ignore the other two end up with a checkout that works in demos and fails in production the moment a different agent shows up.
What
Major PSPs (Stripe, Adyen, Checkout.com) now support both ACP and AP2 agent frameworks
ACP spec added native MCP tool-discovery hook for checkout capability
Raw card numbers an ACP-compliant agent ever sees, via Shared Payment Tokens
Why "We'll Add a Chatbot" Is the Wrong Response
The instinct inside most marketing orgs is to treat agentic commerce as a conversational UI problem and bolt a chatbot onto the existing site. That solves nothing. The agents buying on your customers' behalf are not visiting your site through a browser session with a chat widget. They are pulling structured data through a feed, evaluating your catalog against a query, and attempting a transaction through a protocol your infrastructure either speaks or does not.
The actual gap is upstream of any UI decision. It is whether your product data is structured, current, and machine-readable at all. An agent comparing five vendors on price and availability cannot parse a PDF spec sheet or a marketing paragraph written for a human skimming on their phone. It needs a feed. If that feed does not exist, you do not lose the sale gracefully, you never enter the consideration set.
This is the same failure mode B2B teams already hit with AI purchasing agents evaluating vendor sites: the agent does not "fail" to find you, it simply routes around you toward whoever published clean, structured data it could actually use.
The Readiness Checklist Nobody Has Assigned an Owner To
Ask your team who owns agentic checkout readiness right now. In most organizations, the honest answer is nobody, because it falls between ecommerce, marketing ops, and engineering, and each assumes another team is tracking it.
Agentic
- Publish a structured product feed (price, availability, variants, shipping) an agent can parse without scraping your HTML
- Confirm your payment processor supports ACP's Shared Payment Token flow, not just card-on-file
- Implement AP2 mandate verification so you can prove a human authorized the specific transaction an agent submitted
- Expose your checkout capability through an MCP discovery hook so ACP-aware agents can find it
- Assign a single owner for agent-facing commerce, separate from your human-facing checkout roadmap
- Monitor agent-originated traffic and transactions separately in analytics, since most attribution stacks currently bucket them as bot noise or drop them entirely
Notice what is missing from that list: a rebuilt storefront, a new chatbot, or a six-month platform migration. Every item is an integration or a data quality fix, not a redesign. That is the part most teams get wrong. They wait for a "big agentic commerce initiative" to get funded when the actual work is a feed export and a payment processor configuration change.
Protocol Choice Matters Less Than Being Present
The teams winning the early agentic commerce cycle are not the ones who bet correctly on ACP versus AP2 versus whatever comes next. They are the ones who made their catalog legible to a machine before anyone else in their category bothered. Protocol support from your payment processor is table stakes now that Stripe, Adyen, and Checkout.com all ship it. The differentiator is whether an agent evaluating your product against a competitor's can actually parse what you offer.
Do not wait for a single dominant standard to emerge before acting. ACP and AP2 already interoperate by design, and MCP discovery means new agent frameworks can plug into infrastructure you build today. The cost of acting now is a feed and a processor integration. The cost of waiting is a purchasing channel that closes deals without you in the room.
Start with the audit. Assign the owner. Ship the feed. The agents are already shopping. The only open question is whether they can find you.
Tags
LETSGROW Dev Team
Marketing Technology Experts
Ready to Apply This Insight?
Schedule a strategy call to map these ideas to your architecture, data, and operating model.
Schedule Strategy Call
