Visa stopped picking a winner

Another quiet weekend on the wire, so here is a piece of product design worth reading as a prediction. Visa's Intelligent Commerce Connect lets a merchant accept payments initiated through Trusted Agent Protocol, the Machine Payments Protocol, the Agentic Commerce Protocol and the Universal Commerce Protocol, through one integration on the Visa Acceptance Platform [1].

Visa describes it as network, protocol and token vault agnostic [2]. That phrase is doing a lot of work.

Four protocols, one socket

Visa authored one of those four itself. Trusted Agent Protocol was built with Cloudflare to let agents and merchants authenticate each other through a transaction [3]. A company confident its own standard was going to win would not spend engineering budget terminating three rivals into the same socket.

So read the product as the forecast. Visa is telling merchants that the protocol layer will stay contested long enough that betting on any one of them is a bad use of an integration slot, and that the sensible place to absorb that mess is the acceptance layer, where Visa already sits.

It is the same instinct that produced the x402 Foundation last month, when the Linux Foundation gathered forty founding members including Visa, Mastercard, Stripe, Amex, Adyen, Google and AWS around one open governance body [4]. Everyone joined every table. Nobody is folding.

What it means if you are the one integrating

The pitch also covers catalogue discoverability and offers to handle orchestration and PCI scope on behalf of enablers processing on a merchant's behalf [2]. That second half is the commercially interesting bit, and it is worth being clear-eyed about the trade. You are buying out of a standards fight, and paying for it in dependency.

Read from the rails

I have watched this pattern run in payments twice before, and it goes the same way each time. A format war starts, an intermediary sells an adapter, the adapter gets adopted because nobody wants to bet, and then the adapter becomes the thing everyone is actually integrated with. The standards converge eventually and by then it does not matter, because the switching cost has moved from the protocol to the layer that abstracted it.

That is not an argument against using it. For most merchants this is the correct decision on the merits today. One integration against four moving specifications, with somebody else carrying the PCI scope, is straightforwardly cheaper than picking a horse and re-platforming in eighteen months when it loses.

It is an argument for knowing what you are buying. Write down, now, what it would take to accept an agent payment without this in the path. If the answer is a fortnight, you have bought convenience. If nobody on the team can answer it, you have bought a landlord. The gap between those two answers is worth more than any forecast in this space, and unlike the forecasts it is a question your own engineers can settle this week.

Ten years in payments operations left me suspicious of any integration that is easy to start and undocumented to leave.

Visa built a translator for four languages, one of which it invented. That is not a company expecting its own language to win.