Agentic commerce assumes cards. Most of the world does not

In Poland, cards are 21% of e-commerce transactions. Nearly every agentic payment integration shipped so far assumes a card credential — which makes the protocol debate a local one.

By XAgent Team · 2026-08-15

Almost every agent-payment integration announced this year runs on a card credential. Tokenised card rails, network agent credentials, card-network trust protocols. That is a reasonable place to start if you are building in the United States. It is a strange place to stop if you are selling anywhere else — in Poland, cards account for 21.3% of e-commerce transactions, and the method that accounts for most of the rest could not, until recently, be initiated by an agent at all.

The concentration is easy to miss from inside it

Start with the numbers, because the gap is larger than the debate suggests.

Worldpay's 2026 Global Payments Report, via Global Payments, puts direct card usage at 31% of online spending globally, and in Asia-Pacific much lower: "direct use of cards accounting for just 15% of all online spending and 20% of all in-store spending in 2025, less than half the global average and far below the U.S. usage." Note the methodology — "direct use of cards" excludes cards riding inside digital wallets, so this understates ultimate card dependence rather than overstating it.

In the Netherlands, the Dutch Payments Association — the national scheme body — puts credit cards' "online market share ... stable at around 14%", against iDEAL at around 70%. In Poland, National Bank of Poland data for Q4 2025, as computed by cashless.pl, gives BLIK 70.8% of e-commerce transactions by count and cards 21.3%; by value it is 62.8% and 19%. That computation rests on an explicit assumption the publication states — that four instruments make up 100% of Polish online transactions — so it excludes PayPal, BNPL, cash on delivery and ordinary transfers. Treat it as a shape, not a precise share.

The shape is the point. An agent that can only pay by card, transacting in these markets, is addressing the minority method.

The gap is now being named out loud

On 12 August, PPRO announced a partnership with BLIK, and its framing is the most direct statement of the problem anyone has published: "without support for local payment methods, the agentic economy risks defaulting to a card-centric payment model, limiting access to the payment methods consumers rely on and prefer across many regions."

It also names the asymmetry precisely: "While global card schemes have native features such as tokenisation to streamline the adoption of agentic commerce, PPRO will provide the infrastructure that enables local payment schemes, merchants and PSPs to participate in the AI economy."

Two cautions, because the release is doing more selling than reporting.

It is not shipped. The deliverable is future tense — the project "will introduce" the capability — and the two executives quoted are not describing the same thing. PPRO's chief executive says the partnership has "allowed us to quickly build a framework for agent-initiated transactions in Poland." BLIK's own vice president of the management board says the company is "pleased to explore how BLIK can be used within the evolving agentic commerce ecosystem and to take part in testing solutions that may shape the future of online payments." Explore, take part in testing, may shape. Read the second quote as the status.

And the primacy claim does not hold. The release describes "one of the world's first local payment method implementations for agent-initiated transactions". Alipay AI Pay — an account-based, non-card local method — was already in production well before: Alipay reported in February that AI Pay "exceeded 120 million transactions in the past week", from a product it says launched in 2025. Whatever else is true, agent-initiated payment on a non-card rail at scale is not new; it is just not European.

One more thing worth knowing before framing this as local schemes versus card networks: Mastercard is a shareholder in Polish Payment Standard, which operates BLIK.

The better evidence is Stripe's own roadmap

If you want to test whether agentic commerce is being built card-first, the strongest evidence is not an LPM aggregator's press release. It is what the largest agentic-commerce processor did when it broadened beyond cards.

In March, Stripe expanded its Shared Payment Token support "to enable broader access to network-led agentic payment capabilities, including Mastercard Agent Pay and Visa Intelligent Commerce, as well as buy now, pay later (BNPL) methods such as Affirm and Klarna."

Card network agent credentials, and BNPL. No account-to-account local methods. That is a company with every incentive to support the methods its merchants' customers actually use, reaching first for the two categories that already have a delegated-credential model.

That is the real mechanism, and it is not parochialism. Card schemes had tokenisation and credential-on-file infrastructure before agents existed, so extending them to agent-initiated payments was an increment. A bank-transfer scheme built around a human confirming each payment in a banking app has no equivalent primitive to extend. The agentic path of least resistance runs through cards because cards had already solved delegation for other reasons.

What this means for a merchant selling outside the US

Three practical consequences, none of which require betting on a protocol.

Ask your PSP a question nobody is asking. Not "do you support agentic commerce" but "which of my payment methods can an agent actually initiate?" For most merchants outside the US the honest answer today is: the cards, and possibly BNPL.

Do not read agent-traffic conversion as demand. If agents arriving at your checkout can only complete on the method 21% of your customers use, low completion tells you about the rail, not about the appetite. That is a different diagnosis from the one most dashboards will offer.

Keep the settlement decision out of your storefront. The rail an agent can use will change repeatedly over the next two years, market by market, and none of that changes what you have to do: issue a binding quote, check the purchase against a mandate, create the order, return proof. We have argued this about stablecoin rails and it holds identically for local methods — the merchant-side work is rail-independent, and the integration you regret is the one that hard-codes a rail.

What's next

Watch for the second European local scheme to announce agent support, because one is a partnership and two is a trend. We searched for agent-initiated announcements from iDEAL and Wero, Pix, Swish, Bizum and MB Way and found none — which is what makes the Polish project notable even after its primacy claim is trimmed.

The deeper question is whether local schemes build delegation primitives of their own or accept that agents reach them through an aggregator. That choice will decide whether agentic commerce outside the US looks like the card market with extra steps, or like something that reflects how people in those markets actually pay.

If you would rather be able to take an agent's purchase on whichever rail it arrives with — and produce a record either way — list your store on XAgent and let the open execution market route the settlement.