UCP vs ACP:
agentic commerce.

There are two ways an AI agent can buy something on your behalf, and the difference between them is not really technical. It comes down to one question: when the purchase happens, who owns the checkout and the customer. Everything else is implementation detail.

The shorthand you will hear is ACP versus UCP. ACP is the Agentic Commerce Protocol, backed by OpenAI and Stripe. UCP is the universal commerce, or cart, protocol approach aligned with Shopify and Google. The exact names and boundaries are still settling, so do not over-index on the acronyms. Focus on the philosophy each one encodes, because that is what determines whether you keep your business or rent it.

Let me lay out both, then put them side by side and tell you where Shopify is placing its bet and why I think it is the right one for operators.

Why this is a
philosophy fight,
not a tech one.

Both models solve the same surface problem: a shopper asks an agent for something, the agent needs to find products and complete a purchase. They diverge on where that purchase completes. In one model, the transaction finishes inside the agent. In the other, the agent hands the shopper to the merchant's own surface to check out.

That single design choice cascades into everything that matters to an operator. Merchant of record. Customer data. Retention. Fees. Brand control. So when you evaluate any agentic protocol, skip the architecture diagrams and ask where checkout lands. That tells you who the protocol is really built for.

We have seen this movie before in commerce. Every time a powerful intermediary inserts itself between a brand and its buyer, the same fight plays out. Marketplaces, social commerce, comparison engines: each promised reach, and each quietly accumulated the customer relationship in exchange. The brands that thrived treated those channels as acquisition surfaces while keeping the customer on their own books. The brands that surrendered the relationship became commodity suppliers to someone else's audience. Agentic commerce is the newest round of that same fight, and the protocol you back determines which side of it you are on.

The ACP model:
checkout inside
the agent.

The Agentic Commerce Protocol, the OpenAI and Stripe approach, completes the transaction inside the agent. The shopper never leaves the chat. From a pure convenience standpoint, that is elegant. The friction of jumping to a website disappears, and the agent becomes the storefront.

The cost is who ends up holding the relationship. When checkout lives in the agent, the agent or its payment layer is closer to the merchant of record role, and the customer data and post-purchase relationship are easier for the platform to hold and harder for the merchant to fully own. This is the model that powered ChatGPT Instant Checkout, which reached about a dozen merchants and was pulled within roughly six months. I covered that whole arc in the Instant Checkout postmortem.

The UCP model:
checkout on the
merchant's store.

The universal commerce approach aligned with Shopify and Google flips the last step. Discovery still happens through the agent, so the shopper finds your product in the chat. But checkout completes on the merchant's own surface. You stay merchant of record. The customer is yours. The data is yours.

This is the model behind Shopify Agentic Storefronts, where every store is discoverable by default but the purchase finishes on your store through your checkout. It keeps the agent in the role of introducer rather than owner. For the full breakdown of how that works in practice, see the Agentic Storefronts explainer.

The two models,
side by side,
on what matters.

Treat the fee figures as reported, not official. The 4 percent number tied to certain ChatGPT-originated checkout sales has been reported rather than confirmed as a fixed rate, and ACP fee terms vary by implementation. The point of the table is not the exact percentages. It is the pattern: ACP centralizes ownership in the agent, UCP keeps it with the merchant.

Where Shopify is
betting, and why
I agree with it.

Shopify is betting on the UCP side, and the logic is consistent with everything the platform has done for fifteen years. Shopify makes money when merchants succeed and own their customers, not when a middleman intercepts them. Keeping checkout on the merchant's store is not a technical preference, it is aligned incentives. The platform wins by making you win.

I spent years inside Shopify building the partner program, and then years on the merchant side living with the consequences of who owns the customer. The brands that compound are the ones that keep the relationship. A model that hands that relationship to an agent might spike convenience this quarter, but it quietly weakens the asset that makes a brand worth building. That is why I would build on the model that keeps the customer mine, and discover through every agent I can reach.

Google aligning with the merchant-controlled approach matters too. When the company that has owned discovery for two decades and the company that has owned independent commerce both land on "agent introduces, merchant closes," that is not a coincidence, it is a read on where durable value sits. Discovery is contestable and will be fought over by many agents. The customer relationship, once you own it, is yours to keep. Betting on the layer you can own rather than the layer everyone is fighting to control is just sound business, and it happens to be the layer Shopify is built to defend.

If you take one thing from this comparison, let it be the habit, not the verdict. Protocols will be renamed, merged, deprecated, and relaunched. Run every one of them through the same filter. Where does checkout land. Who is merchant of record. What do I give up to be there. Do that consistently and you will never be the merchant who reorganized a business around a feature that disappeared in six months.

None of this means you should boycott ACP surfaces. If an agent with an in-agent checkout drives real incremental volume and the economics pencil out, take the orders. Be pragmatic. The point is not purity, it is awareness. Know what you are trading when you do it, price the loss of the relationship into your decision, and never let a convenience feature become the foundation your business stands on. Use the surfaces that ask for less control as your primary bet, and treat the ones that ask for more as channels you can walk away from without losing your customer base.

The acronyms will keep shifting. The question will not. Where does checkout land, and who owns the customer after. Answer that, and you will know which agentic surfaces deserve your time.

If you want help mapping these models onto your actual stack, make sure your products are legible to whichever agent asks first, since that work pays off no matter which protocol wins. Then tell me where you are placing your bet and I will tell you what it costs you in control.