Agentic commerce and the private label card: what's shipped, what isn't, and what a merchant should do now

Through 2025 and 2026 the payments industry built the plumbing for AI agents to buy things: Visa Intelligent Commerce and the Trusted Agent Protocol, Mastercard Agent Pay and its agentic tokens, Stripe’s Agentic Commerce Protocol with OpenAI and its Shared Payment Tokens, Google’s AP2 (now under the FIDO Alliance) inside the Universal Commerce Protocol. As of this autumn the plumbing is well ahead of the traffic — only Stripe’s suite is self-serve, the network programs are in pilot or controlled rollout — but the direction is not in doubt.

For a merchant with a private label or co-brand card, the relevant question is narrower than “is agentic commerce real.” It is: when an agent chooses how to pay at my checkout, will my card even be on the list?

What’s already true

Agents default to card-on-file. The first generation of agent checkouts pays with whatever credential the consumer stored. Anything that isn’t a stored card is invisible unless someone wires it in.

BNPL wired itself in. In March 2026 Stripe extended Shared Payment Tokens to Klarna and Affirm (and later Zip), so AI agents can present pay-later options at checkout for any merchant already offering them through Stripe. In May, Google added Affirm and Klarna to checkout in AI Mode and Gemini. Klarna was explicit about why: agents defaulting to card-on-file were “freezing out” alternative payment methods. BNPL providers treated agent checkout as a defensive necessity and moved first.

Issuers are moving, slowly and single-lender. Synchrony has a partnership with OpenAI to embed its private-label cards inside ChatGPT, described as taking months because each retail partner has to be coordinated; it joined the Personal Agent Protocol initiative led by Sierra and Meta as a design partner; and it has integrated CareCredit financing directly into an AI-enabled veterinary practice platform. Synchrony’s own published strategy says the quiet part: private-label cards “need to surface to AI agents not as a single card brand but as a merchant-specific option,” and explanation-heavy products like promo financing are “uniquely exposed” to how they appear inside an AI conversation.

Read those three together. BNPL will be an option in agent checkout by default. Your card will be an option only if your issuer builds the merchant-specific integration and your program is structured so an agent can present it.

What isn’t solved

1. Financing is a decision, not a payment method. Pay-in-4 can be a button. A 0% promotional financing offer requires prequalification, offer selection, disclosure, and consent — inside a conversation. Nobody has shipped a general orchestration layer for that.
2. Compliance in a conversational interface is unmapped. Truth in Lending disclosures, adverse-action notices, and UDAAP exposure in an agent’s recommendation (“I’d suggest the 24-month plan”) have no settled pattern yet, and the question of who is the creditor’s agent when a third-party shopping agent presents the offer is open.
3. Everything is cart-shaped. The protocols assume a SKU and a cart. Considered purchases that are quoted, assisted, or service-based — most of the categories where card programs matter — don’t fit cleanly.
4. There is no waterfall. An agent today gets one BNPL option or a stored card. Presenting the right financing by credit tier and ticket size, in order, is not something any agent surface does.

The ten-point readiness test

A merchant card program is agent-ready if it can answer yes to most of these:

1. Terms, promo financing offers, and eligibility rules are published in a structured, machine-readable form.
2. An authorized agent can call a prequalification endpoint with consumer consent and get a soft-pull decision in seconds.
3. The card is provisioned to digital wallets at approval, so it is usable immediately.
4. Required disclosures and adverse-action notices work inside a conversational flow.
5. You and your issuer have agreed who the creditor’s agent is when a third-party agent presents the offer.
6. The program is registered with the network agent programs through the issuer and processor.
7. Agent-originated applications and purchases are distinguishable in analytics.
8. When an AI assistant is asked “what are the payment options at [brand]?”, the answer is correct and includes the card.
9. Promotional financing is representable as a comparable option alongside pay-in-4 in an agent’s recommendation.
10. Someone has actually run a test purchase through an agent checkout.

Most programs will score two or three. That is not a crisis today; it is the window.

What to do now

• Ask your issuer what they’ve shipped and what’s in their 2027 plan for agent surfaces. Get the answer in writing. If the answer is “nothing yet,” you’ve learned something important about renewal leverage.
• Fix the structured-offer problem now. It costs little, it’s merchant-controlled, and it’s the prerequisite for everything else. It also happens to improve how your financing appears in ordinary AI search answers today.
• Design the prequal-and-consent flow once and use it for web, in-store, and agent. The same soft-pull prequalification that lifts human conversion is what an agent needs.
• Instrument for agent-originated traffic before it exists, so you’ll know when it starts.
• Don’t bet the program on one protocol. Build so the offer can be presented through whichever surface wins; the standards are still consolidating.

The merchants who show up to their issuer in the next twelve months with a ready program and a specific ask will get sequenced first. The ones who wait will inherit whatever default the issuer ships — which, on current evidence, will be a BNPL button.

If you want your program scored against the ten points, that’s the Agent-Ready Credit Program engagement, or start with a 30-minute conversation.

---

Sources and notes. Status of Visa Intelligent Commerce, Mastercard Agent Pay, Stripe Agentic Commerce Suite, and AP2 per public statements and industry reporting through October 2026. Stripe Shared Payment Token support for Klarna and Affirm announced March 3, 2026; Google AI Mode BNPL partnerships announced May 12, 2026. Synchrony’s OpenAI partnership, Personal Agent Protocol participation, and Vetspire integration per Synchrony’s Q3 2026 10-Q and company statements. Quotations from Synchrony’s published agentic-commerce strategy. Regulatory treatment of agent-presented credit is unsettled; nothing here is legal advice.