The Deutsche Bundesbank's stylized flow runs on an instant credit transfer. Its settlement box lists "existing standards" and x402. Its market-structure section predicts that agents, in combination with instant payments in particular, will identify cheaper and faster rails and systematically prioritize cost-efficient means of payment over means with wider margins. Put those together and you have the question of this lesson. When an agent can choose the rail, what is it choosing between?

I take the three rails the course has met and ask each the same questions: how the value moves, when it is final, whether it can be reversed, what it costs, and what the agent learns at steps 8 and 9.

Cards: authorize now, settle later, reverse for months

A card payment is two events separated in time. Stripe's description is as plain as any: at authorization, the bank checks for sufficient funds and holds the amount on the customer's account; at capture, the money moves from the issuing bank to the merchant's account (Stripe). The hold is instant. The movement is not.

Reversal is built in. A dispute, in Stripe's words, occurs when a cardholder questions a payment with their card issuer; the issuer creates a formal dispute on the card network, which immediately reverses the payment and pulls the money, plus dispute fees, back out of the merchant's balance (Stripe). The window is long: Coinbase's comparison table in the x402 whitepaper puts card chargeback exposure at up to 120 days, and the figure is there because the whitepaper is arguing against it.

ACP and UCP both inherit this shape. The scoped token is authorized against a ceiling, captured by the merchant's provider, and disputable afterwards by the cardholder. Step 8 is really two messages here, an authorization now and a settled credit days later, and step 9 is sent on the strength of the first. The rail with the slowest finality has the clearest route back, and that route is the consumer protection the Bundesbank's regulatory box worries agents might leave behind.

Instant credit transfers: ten seconds, final, no route back

The rail the Bundesbank drew. The SEPA Instant Credit Transfer scheme, launched in November 2017, makes funds available in the payee's account within ten seconds of the payment order, 24 hours a day, 365 days a year (European Central Bank). The Eurosystem's settlement service for it, TIPS, settles in real time in central bank money, and the settlement is final and irrevocable (European Central Bank).

Availability is a legal obligation, not a bank's choice. The Instant Payments Regulation of March 13, 2024 requires euro-area providers to offer instant transfers, with the money in the recipient's account within ten seconds, at charges no higher than for a standard credit transfer, with a free service to verify the recipient before the money goes; providers had nine months to receive and 18 months to send (European Parliament).

Read that against the Bundesbank's prediction. A rail that is mandatory, priced at or below the ordinary transfer, and final in seconds is exactly the cost-efficient means an optimizing agent will prioritize. The Bundesbank says so in its central-bank section: financial market infrastructures could face higher volumes because agentic payments prefer cost-effective instant payments. That is a central bank anticipating the agents' step 4 choice landing on its own system.

Finality cuts the other way. There is no chargeback on an instant credit transfer. Once step 7b has happened, the route back is a request to the payee or a liability claim under rules the Bundesbank says are not yet conclusively clarified for agent error. Step 8 on this rail is a single true message: the money is there, for good. And if the agent was wrong, that certainty is the problem. I wrote about the same tension for human payers in The Instant Payments Fraud Paradox. An agent makes it sharper, because the person who could have paused was not there.

Stablecoins: the payment is the settlement

The third rail is the one x402 was built on. A signed authorization moves a stablecoin, usually USDC, from the agent's wallet to the server's, and the transaction is confirmed on chain. The whitepaper's claims for the Base network are settlement in around 200 milliseconds, nominal gas as the only cost, and no reversibility (Coinbase). Its argument is that per-request payments of a fraction of a cent become practical, which is the Bundesbank's micropayment and machine-to-machine case from its retail section.

Two things changed around the rail after the whitepaper. In the United States, the GENIUS Act, signed on July 18, 2025, created the first federal regulatory system for payment stablecoins, requiring 100 percent reserve backing with liquid assets such as US dollars or short-term Treasuries and monthly public disclosure of reserves (White House). And the processors arrived. Stripe accepts x402 payments in USDC on Base, with a minimum of 0.01 USDC, and lands them in the seller's Stripe balance settled in fiat (Stripe). The Bundesbank's phrase, complemented rather than replaced, fits: the stablecoin moves on chain, and the merchant sees dollars in the same balance as its card receipts.

On this rail steps 8, 9 and 10 are one HTTP response. The agent gets the clearest signal of the three rails and the least recourse. The Bundesbank's caution about digital finance applies too: the rail runs on deterministic logic, the agent choosing it does not, and combining the two could, in its words, create new dynamics. My longer view of this rail is in The Stablecoin Settlement Layer Is Here.

Three rails, five questions
CardsInstant credit transferStablecoin (x402)
How value movesAuthorization hold, then capture through the networkBank to bank through the instant payment systemWallet to wallet on chain
When it is finalAt settlement, days after authorizationWithin ten seconds; final and irrevocable in central bank money via TIPSAt on-chain confirmation; sub-second on Base per the whitepaper
Can it be reversedYes, by dispute through the issuer, for monthsNo; recovery is by request or claimNo
What it costsNetwork and processor fees, borne by the merchantAt or below a standard transfer, by regulation in the EUNominal gas; processor fees if a processor is used
Step 8 signal to the merchantAuthorization now, settled credit laterOne message: credited, finalNone separate; the server reads the chain
Route back for the customerChargebackLiability claim or goodwillLiability claim or goodwill
Protocols from this courseACP, UCPThe Bundesbank's stylized flow; UCP through a handlerx402

What the agent optimizes, and who pays for it

The Bundesbank predicts that agents will avoid existing, currently profitable payment networks because those networks cost more than alternatives such as instant payments. The table shows what that trade looks like from the customer's side. The cheaper, faster rails are also the rails without a dispute right. An agent optimizing purely for cost and speed, which is what the Bundesbank says it will do, is also optimizing away the customer's route back.

That is not an argument against instant or stablecoin rails. It is an argument for writing the choice into the mandate rather than leaving it to the optimizer. A limits field that says "instant transfer for merchants on the whitelist, card for anyone else" has priced the dispute right explicitly. A mandate that says "cheapest" has not, and the person who signed it has probably not noticed. Who Pays When the Agent Pays Wrong? sets out how the existing US rules allocate that loss; the short version is that the allocation depends on the rail.

Practical exercise: route the order

Allow 30 minutes. Take the household mandate from Module 1, with its two supermarkets, one pharmacy, €150 weekly ceiling and preference for instant credit transfer, and the step 6 check from the end of Module 4.

  1. Route three cases (10 minutes). For a routine €40 grocery order, the €165 promotion case, and the pharmacy that only takes card (case B), write which rail the agent should use, which execution protocol would carry it, and why. If your limits field does not answer the question, write the line that would.
  2. Write the confirmation chain (10 minutes). For each case, list the messages the agent will receive, in order, and mark the one that means the money is final. Then write the sentence the agent should report to the household at that point. If the honest sentence is "the merchant believes it has been paid," write that.
  3. Write the reversal path (10 minutes). For each case, assume the agent bought the wrong thing. Write who the household calls, what evidence they need from the step 6 record, and whether the rail gives them a route back. If a case has none, note what the mandate should have said to prevent it.

Keep the three cases. Module 6 asks what they look like from the other side of the table: the merchant's, the provider's and the central bank's.

Key takeaways

Six messages, one agent. Steps 7a to 10 are banks talking to banks and a bank talking to a merchant. The agent receives only step 9, and the customer only step 10.

Settlement is complemented, not replaced. The Bundesbank expects change in the interfaces to settlement, not in settlement itself. x402 is the named exception that reaches the money.

A confirmation is not finality. Step 8 is a settlement message, step 9 a commerce message, step 10 the goods. Which one the agent reports as "paid" belongs in the mandate.

Three protocols, two shapes. ACP and UCP carry the mandate inside a checkout session and run on the merchant's existing processor. x402 carries a payment inside an HTTP request and runs on chain. The Bundesbank calls the first two de facto standards; I report that as theirs.

Three rails, three answers. Cards are slow to finality and easy to reverse. Instant transfers are final in ten seconds and cannot be reversed. Stablecoins are final in under a second and cannot be reversed. The agent's cost optimizer prefers the last two.

Write the rail into the mandate. "Cheapest" is a decision to give up the dispute right. Say so, or say which merchants earn it.

Sources and further reading

Next: Module 6, The Paradox. The agent made the payment simpler for the person and more complex for everyone else. Where did the risk go, and who shapes the layers where it now sits?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.

← Previous
ACP, UCP and x402: Three Ways to Carry an Instruction
Next →
Simpler Outside, More Complex Inside