Key Takeaways
- Agentic commerce adds a second customer you have not designed for: the agent that reads structured data and calls an API, not a human you can persuade with photography — Agentic commerce adds a second customer you have not designed for: the agent that reads structured data and calls an API, not a human you can persuade with photography
- It is a data-and-checkout problem, not a redesign — clean feeds, real-time price and stock, machine-readable policies, and a checkout an agent can call
- Two protocol layers do the work: a commerce layer for the checkout (ACP) and a payments layer for provable consent (AP2), with Visa and Mastercard providing acceptance — Two protocol layers do the work: a commerce layer for the checkout (ACP) and a payments layer for provable consent (AP2), with Visa and Mastercard providing acceptance
- Sequence by revenue: expose your highest-velocity, most-reordered SKUs first, because routine reorders are where agents land before anything else — Sequence by revenue: expose your highest-velocity, most-reordered SKUs first, because routine reorders are where agents land before anything else
You now have a second customer
I spent years building commerce for brands where the entire craft was persuading a human: the photography, the reviews, the cart flow, the checkout friction we shaved one field at a time. Agentic commerce does not replace that customer. It adds a second one — an AI agent that reads structured data and calls an API, and is completely indifferent to your hero image. The shopper sets the intent ("reorder what fits my machine, here by Friday"); the agent does the browsing and the buying. The question for a commerce leader is no longer only "will this convert a person," but "can a machine parse and purchase this at all."
It is a data problem, not a redesign
The instinct when a new channel appears is to commission a redesign. Resist it. Being agent-ready is not a new storefront; it is an API surface beside the one you have. Four things carry it: a clean product feed with stable identifiers, real-time price and stock served programmatically rather than only rendered on a page, machine-readable shipping and returns and tax, and a checkout an agent can call. If you have invested in a customer data platform and disciplined product data, you are closer than you think. If you have not, this is one more reason the data foundation is the work.
Two protocol layers, and why the split matters
The standards landed fast, and the headlines make them sound like a format war. They are not. There are two layers doing two jobs. A commerce layer — the Agentic Commerce Protocol, from OpenAI and Stripe — handles the checkout: how an agent discovers what it needs and submits an order. A payments layer — the Agent Payments Protocol, originated by Google and now governed by the FIDO Alliance — carries a cryptographic mandate proving the shopper authorized that exact spend. Visa and Mastercard sit underneath with acceptance, and their protocol-agnostic on-ramps mean you can support several agent standards through one integration rather than betting on a single winner. For a leader, the takeaway is simple: integrate the commerce layer so agents can reach you, then wire payments so the spend is provable.
The pricing question gets sharper, not softer
An agent comparing options across merchants in milliseconds is a more efficient price-discoverer than any human shopper, which raises the stakes on how you price. This is where agentic commerce and dynamic pricing meet: if agents are reading your price programmatically and weighing it against alternatives, a stale or poorly-constrained price is exposed faster and more often. The discipline I would bring to a dynamic pricing system — clean signals, tight constraints, a human override — is exactly the discipline the agent channel rewards.
Sequence by revenue, test the failure paths
Do not try to make the whole catalog agent-ready at once. Start where the agent purchases will actually be: your highest-velocity, most-reordered SKUs. Routine reorders are the first behavior agents pick up, so that subset is where the early revenue and the early learning are. It also gives you a contained surface to test the things that break — a stale price the agent committed to, an item that sold out mid-transaction, a missing consent mandate, a duplicate submission — before you widen coverage. An agent hits those failure paths far more often than a human, because it transacts faster and without the visual cues a person uses to bail.
The leader's call
Treat the agent as a customer you have not designed for yet, and treat readiness as a sequenced investment rather than a moonshot. The work is structured data and a callable checkout wired to provable consent — cheap to start, expensive to retrofit. The brands that are parseable and buyable by agents will quietly win the routine purchases first, and routine purchases are a large share of commerce. I would not bet the company on it this year. I would absolutely make sure my best SKUs were ready, because being invisible to the agent channel is a decision too, just not one anyone meant to make.
Frequently asked questions
What is agent-ready commerce?
It is the state of being purchasable by an AI agent, not just a human. A shopper tells an assistant to reorder something or find an option under a budget; the agent reads your product data, decides, and completes checkout through a commerce protocol. Agent-ready means you expose the structured data and the callable checkout an agent needs — a clean product feed, real-time price and stock, machine-readable policies, and a checkout endpoint that speaks the protocol. It is an API surface added beside your storefront, not a replacement for it.
Is agentic commerce actually worth preparing for now, or is it hype?
The protocols and the card-network rails both shipped in early 2026, which is the line between concept and channel. You do not need to bet the business, but you do need to be parseable, because the cost of readiness is low and the cost of being invisible to agents is losing the routine-reorder purchases to competitors who are ready. I would treat it the way I treated mobile commerce in its early years: not an emergency, but a deliberate, sequenced investment that is cheap now and expensive to retrofit later.
What does my team actually have to build?
Four things, none of which is a redesign. A clean product feed with stable identifiers. Real-time price and stock served programmatically, not just rendered on a page. Machine-readable shipping, returns, and tax policies. And a checkout an agent can call, wired to a payments layer that only clears with provable consent. On a hosted platform, most of this is arriving as configuration rather than custom code, so for many teams the lift is smaller than it sounds.
Which protocols matter — ACP, AP2, Visa, Mastercard?
They are layers, not competitors. The Agentic Commerce Protocol (OpenAI and Stripe) handles the checkout interaction. The Agent Payments Protocol (Google, now under the FIDO Alliance) carries the cryptographic mandate that proves the shopper authorized the spend. Visa’s Intelligent Commerce Connect and Mastercard’s Agent Pay provide acceptance on card rails, and their protocol-agnostic on-ramps let you support several agent standards through one integration. Adopt the commerce layer first so agents can reach you, then wire payments through your processor.
How is this different from the chatbots we already have?
A chatbot recommends and hands the shopper to a website. An agent transacts. That single difference changes what you optimize for: the persuasion surface (images, reviews, copy) shrinks in importance for the agent path, and the data surface (feeds, real-time signals, a callable checkout) becomes the thing that determines whether the agent can buy from you at all. You still serve the human storefront; you are adding a machine-readable path next to it.
Where should I start if I only do one thing?
Make your highest-velocity, most-reordered SKUs cleanly parseable and buyable: stable IDs, real-time price and stock, and a callable checkout for that subset. Routine reorders are where agents land first, so that is where the early revenue is, and it gives you a contained surface to test the failure paths — stale price, out-of-stock mid-transaction, invalid mandate, duplicate submission — before you expand catalog coverage.
Ready to Transform Your AI Strategy?
Get personalized guidance from someone who's led AI initiatives at Adidas, Sweetgreen, and 50+ Fortune 500 projects.
