Google Shipped Its Second Protocol in Four Months. The Retailer Still Keeps the Chargeback.

Google published a payments protocol in September and a commerce protocol in January. The second one is where the money is — and the retailer still eats the dispute.

Share
Google Shipped Its Second Protocol in Four Months. The Retailer Still Keeps the Chargeback.

In September Google published a protocol for how an agent pays. In January it published one for how an agent shops. Only one of those questions is worth owning.

On 11 January, at the National Retail Federation's annual show, Sundar Pichai announced the Universal Commerce Protocol. He described it as designed for the era of agentic commerce, open and agnostic, and available starting that day. Shopify, Etsy, Wayfair, Target and Walmart built it with Google. More than twenty other companies have endorsed it.

It is compatible, he said, with Agent2Agent, with the Agent Payments Protocol, and with the Model Context Protocol.

Read that sentence again, because it contains the thing worth writing about. Google published the Agent Payments Protocol in September. Four months later it has published a second protocol and told us the two are compatible with each other.

Which raises an obvious question that nobody seems to have asked out loud: if the first one solved agentic commerce, what is the second one for?

Pay versus shop

The answer is that they solve different problems, and the second problem is the valuable one.

AP2 is a payments protocol. Its contribution is the mandate structure — an intent mandate setting the conditions under which an agent may buy, a cart mandate fixing specific items and prices, a payment mandate authorising the final transaction. It is a careful answer to a narrow question: how do you prove what a human actually agreed to before an agent spent their money.

UCP is not that. UCP is about discovery, decision and checkout — the whole path from a customer wanting something to a customer having bought it. Pichai's own framing covers retailers and customers across discovery, decision, and beyond.

Those are not two parts of one stack. They are two different businesses, and anyone who has worked in payments knows which one has historically been worth more.

The payment is the last two seconds of a commercial relationship and the least differentiated part of it. Discovery is where the relationship forms, where the margin is set, and where the customer decides whether a merchant exists at all. Google did not publish a payments protocol twice. It published a payments protocol, and then moved up the stack to where it has always made its money.

That is not a criticism. It is the single most rational move available to the company, and it should be read as a signal about where the value in agentic commerce is expected to sit. The people building payment rails for agents should notice that the largest participant in this market built one, shipped it, and then went somewhere else.

The sentence that matters to us

Now the part that an issuer or processor should underline.

The retailer, Pichai said, remains the merchant of record.

That is one clause, it sounds like a reassurance to retailers, and it is the whole liability architecture of the product.

Merchant of record is not a label. It determines who holds the acquiring relationship, who appears on the cardholder's statement, who receives the chargeback, who assembles the representment evidence, who carries the fraud loss, who owes the sales tax, and who owns the transaction data. Every hard and expensive thing in a commercial transaction attaches to the merchant of record.

So Google has built the layer that decides what a customer sees, how an offer is personalised, and where a purchase begins — and has explicitly declined to become the party that eats a dispute.

I want to be fair about this, because the alternative would be worse. A world where Google became merchant of record for everything it surfaced would concentrate commerce in a way nobody should want, and retailers would lose the customer relationship entirely. Keeping the retailer as merchant of record is the right outcome for retailers and it is what they would have negotiated for.

But the shape should be named accurately. The layer with the influence has none of the liability. The party with the liability has less influence over the sale than it had a year ago. That gap is where the next five years of argument in this industry will happen, and it opened on 11 January without anyone objecting.

Open, agnostic, and whose

In December I wrote that an open-source licence is not a governance structure, and that the agentic payment protocols had confused the two. UCP arrives with the same ambiguity, and it is worth being precise rather than cynical about it.

Pichai calls UCP open and agnostic. Google's own announcement describes it as an open-source agentic commerce protocol. What the announcement does not say — and I have read it looking for this — is that UCP has been given to anybody. There is no standards body named. There is no foundation. There is a specification published by a company, five retailers who built it with that company, and more than twenty endorsers.

That is adoption. Adoption is a real achievement and it is not governance. The difference matters for exactly one reason: when the interests of the publisher and the interests of the adopters diverge, a governance structure decides and a licence does not.

I hold this lightly, because the right move may still be coming. Protocols have been donated before, and a company that has just launched something is not obliged to hand it over in the same week. The question to carry forward is simply whether UCP ends up with a neutral steward, and the answer will tell you whether this is a standard or a product.

Meanwhile, count the protocols. ACP from OpenAI and Stripe. AP2 from Google. x402 from Coinbase. MCP, donated. And now UCP, also from Google. In October I wrote that three protocols had arrived in thirteen days and did not agree on what an agent is. The count has gone up and the agreement has not.

Who built it, and who did not

The partner list deserves the same reading I gave the others.

Shopify. Etsy. Wayfair. Target. Walmart. Google.

Five retailers and a search company. No bank. No card network. No acquirer. No processor. No payment service provider. Nobody who moves the money, carries the settlement risk, or answers the phone when a transaction is disputed.

A protocol that specifies how commerce is discovered and how checkout begins was designed entirely by the demand side of the market. The supply side of payments — the institutions that will have to authorise these transactions, settle them, and adjudicate them when they go wrong — was not in the room.

That is the same observation I made about the sixty companies behind AP2 and I am going to keep making it, because the pattern is now four protocols deep. The people who will inherit the operational consequences of these specifications are consistently absent from writing them, and the reason is boring rather than sinister: standards get built by whoever shows up, and nobody from this side of the business has shown up.

One more detail worth noticing, because it reads as a small convenience and is not. A customer can complete a purchase with a few taps via Google Pay, without leaving the conversation.

The retailer keeps merchant of record. The wallet is Google's. That combination — somebody else's wallet in front of your merchant-of-record obligations — is the arrangement I wrote about in November when PayPal brought its own buyer protections to agentic checkout. It concentrates the customer relationship in the wallet and leaves the liability with the merchant, and it is now the default shape of agentic commerce rather than one company's strategy.

Why now, in one number

There is a figure in Pichai's remarks that has nothing to do with commerce and explains the timing better than anything in the protocol.

Google's API processed 8.3 trillion tokens in December 2024, and more than 90 trillion a year later.

Roughly eleven times, in twelve months. That is the demand curve underneath all of this. Not a standards breakthrough, not a payments innovation — a volume of machine inference that did not exist two years ago and now has to be pointed at something that makes money. Commerce is the obvious thing to point it at.

Whenever somebody asks why all of this is happening at once, that is the answer. The capability arrived first and the business model is being retrofitted to it at speed. Which is exactly the condition under which important structural questions — like who pays when it goes wrong — get deferred.

From a Panamá desk

I have just moved to Panamá, and one of the first things you notice from a smaller market is how differently a launch like this reads depending on where you are standing.

In a large market, UCP is a strategic question: do we integrate, on what terms, and what does it do to our margin. In a market this size it is not a question at all in the short term, because none of the five retailers who built it operate here, and the buy button is coming to Google surfaces whose commerce features arrive in this region late or not at all.

So the honest near-term answer for a Panamanian institution is: nothing happens to you in 2026 because of this.

And the honest medium-term answer is the one that matters. What arrives here is not UCP. What arrives is the consequence of UCP — cardholders in this market buying from merchants in other markets through an agentic surface, on a card we issued, under a merchant-of-record arrangement nobody told us about, with a dispute that lands in our queue carrying no field that says any of this happened.

That is the same conclusion I reached about the network credentials in December, and it has not improved. The category reaches us as a change in the composition of our cross-border card-absent traffic, with no label attached, and the first evidence is a phone call.

The cheap preparation is the same as it was: ask your processor, in writing, what in the message would distinguish an agent-initiated transaction, whether it is stored, and whether you can report on it. Keep the answer. It costs an email and it is the difference between noticing a trend and being told about one.

The second thing I would do, and this one is free, is read the specification. UCP is published. Whatever its governance turns out to be, the document is available, and an institution that has read it will be a year ahead of one that waits for a vendor summary. Nobody in this region needs permission to read a specification. We have simply not been in the habit.

My judgment

The interesting thing about January is not that there is a fourth protocol. It is where the fourth protocol was aimed.

Google built a payments protocol in September, and then built a commerce protocol in January, and the second one is the one it launched on a retail stage with five named retailers and a buy button. That is a company telling you, in its own sequencing, that the money in agentic commerce is in discovery rather than in settlement.

I think that is correct, and I think the payments industry has been slow to absorb it. We have spent eighteen months arguing about credentials, mandates, tokens and liability — all of which matter, all of which are our responsibility, and none of which determine who captures the value. The party that decides which merchant a customer sees has always captured more than the party that moves the money, in every channel, in every decade. Agentic commerce is not going to be the exception.

My second judgment is narrower and more confident. Merchant of record is going to be the most contested phrase in commerce over the next three years. Every agentic surface will want the influence without the liability, and every retailer will want the relationship without being the only party exposed. The settlement of that argument — contract by contract, and eventually rule by rule — is where this industry's next decade of margin is decided, and almost nobody outside the acquiring business is paying attention to it.

The third is a prediction I will be glad to lose. UCP will not be donated to a neutral body within a year. If it is, it will be because an adopter with real leverage demanded it as a condition, not because the publisher offered it. I have now watched this sequence four times and the pattern has not broken once.

Three questions for the people who own this

1. For the agentic surfaces your cardholders already use, who is the merchant of record — and does your dispute process know the difference between the merchant of record, the retailer, and the surface the customer thinks they bought from?

2. Has anyone at your institution read a protocol specification in full this year? Not a vendor summary. The document. If not, that is a half-day of somebody's time and the cheapest competitive advantage available to you.

3. When a transaction arrives through a wallet you do not control, at a merchant of record your customer has never heard of, initiated by an agent nobody declared — which of those three facts does your authorization message currently carry? If the answer is none, what would it take to carry one?

Reply to this email with a line or two. The sharpest answers will shape the next issue.

Sources

Google, 11 January 2026 — Sundar Pichai's remarks at the National Retail Federation, published on Google's blog. Source for the date and venue; the announcement of the Universal Commerce Protocol as designed for the era of agentic commerce, serving retailers and customers across discovery, decision and beyond; the description of it as open and agnostic, with the announcement's own standfirst calling it an open-source agentic commerce protocol; that Shopify, Etsy, Wayfair, Target and Walmart built it with Google and more than twenty others have endorsed it; that it is compatible with Agent2Agent, the Agent Payments Protocol and the Model Context Protocol; that it is available starting that day, with native checkout first and a buy button coming soon to Google surfaces including AI Mode in Search and Gemini; that retailers can personalise offers at checkout, including a new-member price, instant loyalty enrolment and deals based on past purchases; that customers can complete a purchase with a few taps via Google Pay without leaving the conversation; that the retailer remains the merchant of record; and the figures that Google's API processed 8.3 trillion tokens in December 2024 and more than 90 trillion a year later, and that the Shopping Graph holds over 50 billion product listings.

On what the announcement does not say. It names no standards body or foundation for UCP and does not state that the protocol has been donated to one. It gives no UCP-specific volume, merchant or revenue figures, no pricing, and no date for the buy button beyond "soon." It does not use the acronym AP2, so the identification of the Agent Payments Protocol with AP2 is mine, based on Google's September publication of a protocol by that name. Every statement here about governance is an observation about what the announcement omits, not a claim that Google has refused anything.

A note on what is mine. That AP2 and UCP answer different questions, and that the second — discovery — is the commercially valuable one; that the sequencing of the two publications is itself a signal about where Google expects the value to sit; that "the retailer remains the merchant of record" is a liability architecture rather than a reassurance, and separates influence from exposure; that a protocol built by five retailers and a search company reflects the absence of the payments supply side from standards work; that merchant of record will be the most contested phrase in commerce over the next three years; and the prediction that UCP will not be donated to a neutral body within a year — those are arguments and forecasts, not findings. The description of what merchant of record determines is standard industry practice and comes from my own work. The characterisation of AP2's mandate structure comes from Google's September publication, not from this announcement. Nothing here reflects the position, portfolio, clients or vendors of my employer.