Agentic payments will be won in B2B first — banks blocking consumer agents are defending the wrong front

Opinion · Finlane Editorial · Outlook & Opinion

On Tuesday's earnings call, Jamie Dimon told AI agents to stay out of Chase, and Citi soon said the same. Neither bank will let a third-party agent log in with a customer's credentials to pay a bill or move money. The same week, Visa, Bill and Billtrust switched on what they call the first live pilot of agents paying approved supplier invoices end to end.

One door closed and another opened. We think the open one matters more: agentic payments will scale in B2B well before they scale for consumers.

Why business payments go first

Look at how the Visa pilot works. The agent doesn't get the keys to the account. It's registered and authenticated through Visa's Trusted Agent Protocol, uses tokenised card details and can only pay what the buyer has already approved. That isn't a leap into the unknown. It's accounts-payable automation with one more step taken out.

Business payments already have the things agents need: approval workflows, spending limits, purchase orders and audit trails. Even the question that worries JPMorgan most, who pays when an agent gets tricked, has old answers in B2B: delegated authority, dual control and limits per user. Consumers have none of that scaffolding.

Why the consumer block won't last

We've seen this film before. Banks blocked screen-scraping a decade ago, and it ended with regulated APIs, not with the aggregators going away. Handing an agent your password really is a bad idea, so the block is fair for now. But as Simon Taylor puts it, customers who want an agent will get one. The only question is whether the bank provides safe access or watches someone else own the relationship. Mercury already lets an agent prepare a payment for the customer to approve in the app. That's the model the big banks are refusing, for now.

European banks start from a better place than they might think. PSD2 already made them build access interfaces with strong customer authentication. Adding agent-initiated payments with an approval step is a smaller job than the one US banks face.

The case against

Fraud, not technology, is what holds this back. Authorised push payment scams are already among the hardest losses to stop, and an agent that can be talked into things is a new way in. A couple of headline agent frauds could make regulators slow everything down, B2B included. And corporate finance teams are cautious buyers. A pilot with selected suppliers is a long way from an AP department trusting agents with its whole supplier base. We could be wrong on timing. We don't think we're wrong on direction.

What we'd do

• If you sell payments or AP software, build for delegated authority now: agent identities, per-agent limits, human approval steps and full audit logs. We expect these to show up in RFPs within 12–18 months.
• If you run a bank, offer agents a permissioned way in, read-only first and then payments with step-up approval in your app, before platforms route around you. Treat it as the next open-banking interface, not a security exception.
• If you sell fraud or identity tools, "know your agent" is becoming a category: which agent acts for which customer, under which mandate.
• If you invest, be wary of consumer-agent payment start-ups that depend on customers sharing passwords. Back the plumbing every side will need: tokens, mandates and agent identity.

What we're watching

• Whether Visa and Mastercard publish liability rules for agent-initiated payments as pilots move into production.
• The first European bank to open an agent-access API, most likely built on its PSD2 interface.
• Whether JPMorgan or Citi put a date on letting outside agents in, or ship their own first.

Comparing payments, open banking or fraud providers for this? Finlane lists them here.

Sources: Fintech Brainfood · Finextra