There is an infrastructure problem at the heart of agentic commerce that most coverage of AI shopping assistants has ignored. When an autonomous AI agent purchases something on behalf of a consumer, it defaults to a stored card — a credit or debit card already on file. That is the only payment method most agent architectures can handle. Every alternative payment option — buy now pay later, bank transfers, digital wallets, instalment plans — gets silently excluded from the transaction.
Klarna and Stripe have moved to close that gap. Klarna’s flexible payment options will now be supported within AI agent-driven checkout flows through Stripe’s Shared Payment Tokens (SPTs), a mechanism designed to let autonomous systems initiate purchases using a customer’s preferred payment method without the agent ever accessing the underlying payment credentials.
For US merchants already running Klarna through Stripe, no additional integration is required. The capability layers into existing infrastructure. But the significance extends well beyond a single payment provider adding support for a new transaction channel.
What Shared Payment Tokens actually solve
The technical constraint is straightforward but consequential. AI shopping agents operate on behalf of a consumer but are not the consumer. They cannot authenticate through a bank’s 3D Secure flow, cannot complete a BNPL credit check, and cannot interact with payment interfaces designed for human users. Standard card-on-file transactions work because the token is pre-authorised and requires no interactive steps. Everything else breaks.
Stripe’s SPTs create a permissioned layer between the consumer’s actual payment methods and the agent executing the purchase. The agent can reference a specific payment option — including Klarna’s interest-free credit — without seeing card numbers, bank details, or credit terms. The authorisation happens through the token infrastructure rather than through user-facing authentication flows.
| Payment Scenario | Without SPTs | With SPTs |
|---|---|---|
| Card-on-file purchase | Works — agent uses stored card token directly | Works — no change to existing flow |
| Buy now, pay later (Klarna) | Blocked — agent cannot complete interactive credit check or redirect flow | Supported — agent references Klarna token, authorisation handled through permissioned layer |
| Bank transfer / open banking | Blocked — requires bank authentication the agent cannot perform | Future potential — SPT architecture extensible to additional payment methods |
| Consumer payment preference | Overridden — agent defaults to whatever card is stored regardless of user preference | Preserved — agent can select from consumer’s full range of authorised payment methods |
This is a meaningful shift. When AI agents can only use stored cards, consumers lose the financial flexibility they have come to expect from modern checkout experiences. For a consumer who routinely uses interest-free instalments for larger purchases, having an AI agent bypass that option and charge the full amount to a debit card is not just inconvenient — it changes the economics of the transaction from the buyer’s perspective.
Why this matters for payment infrastructure
The deeper issue is about who controls the checkout experience in a world where purchasing decisions are increasingly delegated to software. Today, the number of transactions executed by autonomous agents is small. Within a few years, as AI assistants from Google, Apple, Amazon, and standalone startups mature, a significant share of e-commerce volume will flow through automated purchasing systems.
If the infrastructure powering those systems only supports card-on-file transactions, then every alternative payment method — BNPL, real-time bank payments, digital wallets, instalment plans — risks being structurally excluded from the fastest-growing transaction channel in e-commerce. That is an existential threat for companies like Klarna whose entire business model depends on being present at the point of purchase.
Stripe’s position in this shift is worth noting. By building the token infrastructure that connects payment providers to AI agents, Stripe is positioning itself as the essential middleware layer for agentic commerce. The same way Stripe became the default payment infrastructure for internet businesses in the 2010s, SPTs represent a bid to become the default payment infrastructure for AI-driven commerce in the 2030s. Every payment provider that wants access to agent-executed transactions will likely need to integrate through this or a similar token layer.
The merchant perspective
For merchants, the immediate benefit is conversion protection. BNPL options consistently lift checkout conversion rates — Klarna’s own data suggests double-digit improvements in many categories. If AI agents cannot offer BNPL at checkout, merchants lose that conversion lift on every agent-executed transaction. As the share of agent-driven purchases grows, that revenue impact compounds.
The integration is designed to be invisible from the merchant’s side. If a business already accepts Klarna through Stripe, agentic transactions flow through the same infrastructure with no additional development work. This frictionless approach is deliberate — Stripe and Klarna are betting that the payment providers who require zero merchant effort to support agentic commerce will capture the market before competitors who demand new integrations.
What this signals for community banking and financial services
For financial institutions watching the agentic commerce trend, the Klarna-Stripe partnership illustrates a pattern that will repeat across financial services. When AI agents begin managing transactions on behalf of consumers, every financial product that requires interactive human authentication faces a compatibility problem. Mortgage applications, insurance quotes, investment trades, loan originations — any process designed around a human clicking through screens and providing real-time consent will need to be re-architected for a world where software acts on the consumer’s behalf.
The institutions that build token-based, API-first interfaces for their financial products will remain accessible to AI agents. Those that rely on legacy web forms and manual authentication flows will find themselves increasingly invisible to the automated systems through which a growing share of financial decisions are made.
The checkout is just the beginning. The infrastructure being built today for agentic payments will define how AI systems interact with the entire financial services stack for the next decade.