Walmart Gave Agentic Commerce a Catalogue. PayPal Gave It Something Rarer.
The first dispute mechanism in agentic commerce did not come from a card network. It came from a wallet.
Two weeks in October. The second announcement contained the first dispute mechanism anyone has shipped, and it did not come from a card network.
Agentic commerce has spent the year acquiring infrastructure. Credentials in April, when Mastercard and Visa opened their rails within a day of each other. Protocols in September, when Google, Coinbase and OpenAI each published a standard inside thirteen days.
What it did not have was anything to buy, or anything to do when a purchase went wrong.
In two weeks of October it got both, and the second one is far more important than the coverage suggested.
14 October. Walmart announced a partnership with OpenAI that will let customers and Sam's Club members shop its catalogue inside ChatGPT, with Instant Checkout as the purchase method. The largest retailer in the world, arriving in the chat window.
28 October. PayPal announced it is adopting the Agentic Commerce Protocol, bringing its merchant network into ChatGPT and processing payments for merchants using Instant Checkout — and, in a clause almost nobody quoted, bringing with it PayPal's buyer and seller protections, and post-purchase services including tracking and dispute resolution.
That second announcement is the first time any participant in agentic commerce has shipped a mechanism for handling a purchase that goes wrong.
The catalogue problem, briefly
Walmart first, because it is the simpler of the two.
Until October, the agentic commerce proposition had a hole in the middle of it: there was very little to buy. Etsy sellers were live in Instant Checkout, Shopify merchants were listed as coming, and a shopping experience built on a partial catalogue is not a shopping experience. An agent that cannot find the thing you want is a worse search engine.
Walmart closes a large part of that. Doug McMillon framed it against the status quo — that for many years ecommerce has consisted of a search bar and a long list of item responses — and Sam Altman's line was characteristically modest, about making everyday purchases a little simpler.
Two things in that release are worth noticing that the coverage did not.
The first is how little it commits to. The feature is described as coming soon. There is nothing about account linking, nothing about how payment is handled, nothing about what data moves between the two companies. For an announcement of this size, the operational content is close to zero. That is not unusual for a partnership announced at the intent stage, but it means there is nothing here to assess yet.
The second is that McMillon referred, in the same release, to Walmart's path forward with Sparky — its own assistant. A retailer announcing that it will sell through somebody else's agent while keeping its own assistant in the frame is hedging, visibly, in the press release itself. I would not read too much into one phrase. I would read something into it.
The protection is the real news
Now PayPal, and the clause that matters.
PayPal is doing several things at once. It is adopting ACP. It is processing card payments for Instant Checkout merchants through the delegated payments API. Its ACP server handles merchant routing, payment validation and orchestration, so merchants can list in ChatGPT without building individual integrations. Merchant catalogues arrive in 2026 and sign-ups are open now.
All of that is a straightforward and well-executed acquirer play. PayPal is positioning itself as the aggregation layer for agent commerce — the thing that a thousand merchants plug into instead of each building their own connection. That is the business PayPal has been in for twenty-five years, pointed at a new surface.
But the clause I keep coming back to is the one about protections.
Buyers using PayPal in Instant Checkout get PayPal's buyer and seller protections, plus post-purchase tracking and dispute resolution.
Read that against everything else shipped this year. Visa's commerce signals help to manage disputes. Mastercard's intent data can help prevent or resolve them. Google's mandate chain provides a basis for accountability. Three careful hedges, no mechanism.
PayPal just attached an actual one. Not a framework, not an audit trail — an existing protection scheme with a claims process, applied to agent purchases.
Why a wallet could do it and a network could not
The reason PayPal could ship this is worth stating precisely, because it predicts how the rest of this is going to go.
PayPal's buyer protection is not a card network rule. It is a contractual scheme PayPal operates itself, funded by PayPal, adjudicated by PayPal, extended or narrowed by PayPal. There is no committee. There are no thousands of member institutions whose losses are being reallocated. PayPal can decide on a Tuesday that a category of purchase is covered, and on Tuesday it is covered.
A card network cannot do that. A liability rule on an open four-party scheme is a reallocation of losses among institutions the network does not employ, in a hundred countries, each of which gets a say and many of which will object because it costs them money. That is a consultation, a mandate, an effective date and a compliance programme — not an announcement.
So the first dispute mechanism in agentic commerce did not arrive from the card rulebook. It arrived from outside it, imported from a private scheme that already had one and could extend it alone.
I think that is the structurally important fact of October, and it points somewhere uncomfortable.
The agentic liability gap may not get closed by the networks at all. It may get routed around by wallets.
Follow that through. If protections for agent purchases are available from wallets and not from cards, agentic volume concentrates in wallets. The card underneath becomes a funding instrument — the thing that tops up the wallet, not the thing the customer transacts with. And when something goes wrong, the dispute is raised with the wallet, adjudicated by the wallet, and resolved under the wallet's rules.
For an issuer, that has one pleasant consequence and one serious one.
The pleasant one: fewer disputes. If PayPal absorbs the agent-error argument, it does not reach our queue.
The serious one: we stop seeing it. The dispute is where an issuer learns what is happening in its portfolio — which merchants are failing, which patterns are emerging, where fraud is moving. Lose the dispute and you lose the signal, and you are left funding a channel whose risk profile is visible to somebody else and not to you.
That is disintermediation at the exact moment the customer relationship matters most: when they have a problem.
From a Mexico City desk
Neither announcement reaches this region at launch. Walmart's catalogue is a US proposition. PayPal's merchant catalogues arrive in ChatGPT during 2026, with no stated date.
But PayPal is the one to watch here, for a reason specific to how this region actually buys.
PayPal has meaningful penetration across Latin America as a cross-border payment method, and a large part of why people use it for international purchases is precisely the buyer protection. When somebody here buys from a foreign merchant they have never heard of, PayPal is often chosen not because it is cheaper but because there is somebody to complain to who is not the merchant.
That existing habit is the distribution channel for agentic commerce in this region. Not cards first, as with most previous waves — PayPal first, carrying its own protections, aimed at buyers who already trust it for exactly this kind of purchase.
Which means the issuer consequence I described above may arrive here faster than it does in markets where domestic card acceptance is stronger. Our cardholders are already accustomed to routing their riskiest purchases through a wallet. Agent purchases are, by construction, their riskiest purchases.
The practical question for an issuer here is therefore not whether to support agentic commerce. It is what proportion of cross-border card-absent volume already funds a wallet rather than paying a merchant directly — and whether anyone is tracking whether that proportion is moving.
That number exists in your authorisation data today. Nobody I know reports it.
My judgment
October gave agentic commerce a catalogue and a protection, and the second is the one that will shape the market.
The catalogue was always going to arrive. Retailers want distribution, and a large retailer appearing in a large chat interface was a matter of when.
The protection was not inevitable, and where it came from matters more than that it came. It came from a private scheme that could act alone — and that tells you the sequence for the rest of this. Private schemes will move first, because they can. Card networks will move later, through rules processes, under competitive pressure created by the private schemes having already moved.
My prediction, offered so it can be checked: the next agentic protection announcement will also come from somewhere that does not need permission — another wallet, a closed-loop network, a large platform underwriting its own marketplace. Not from Visa or Mastercard, not this year.
And for those of us issuing the cards underneath: the risk is not that we end up carrying the agent-error losses. It is that we end up carrying none of them, see none of the disputes, and learn nothing about a channel our customers are using heavily.
Three questions for the people who own this
1. What share of your cross-border card-absent volume funds a wallet rather than paying a merchant directly? If that share is growing, your view of what your customers are actually buying is shrinking.
2. If a wallet adjudicates an agent-purchase dispute and never raises it with you, what did your institution learn? And is there anything in your agreements that would let you see it?
3. PayPal extended an existing protection scheme to a new transaction type by deciding to. Does your institution have any protection scheme it could extend the same way — or is every protection your customers have granted by somebody else's rulebook?
Reply to this email with a line or two. The sharpest answers will shape the next issue.
Sources
Walmart corporate newsroom, 14 October 2025 — the partnership with OpenAI enabling Walmart customers and Sam's Club members to shop through ChatGPT using Instant Checkout, described as coming soon; quotes from Doug McMillon, President and CEO of Walmart Inc., including the reference to the path forward with Sparky, and from Sam Altman of OpenAI. The release gives no detail on account linking, payment handling or data sharing. Link
PayPal newsroom, 28 October 2025 — PayPal adopts the Agentic Commerce Protocol, processes payments for merchants using OpenAI Instant Checkout through the delegated payments API, and connects its global merchant network to OpenAI; its ACP server handles merchant routing, payment validation and orchestration; buyers paying with the PayPal wallet receive PayPal's buyer and seller protections and post-purchase services including tracking and dispute resolution; merchant product catalogues arrive in ChatGPT in 2026 with sign-ups open now; quote from Alex Chriss, President and CEO of PayPal. Link
Both are company releases. The argument that a private protection scheme can extend unilaterally where an open four-party network cannot, and the consequences for issuer visibility, are mine. Neither company makes any such claim.