The Complexity of the Modern Payments Stack
3 BIG IDEAS FROM THE PODCAST
Last month at Money20/20, I had the pleasure of sitting down with Marqeta CEO Mike Milotich, expecting to talk about agentic AI and payments, which (spoiler alert) we absolutely did.
But what also emerged was a larger narrative about how the payment stack is reorganizing itself. The front end keeps getting simpler. The back end keeps absorbing more complexity. And the true center of gravity is shifting toward the payment credential itself, not whatever happens to trigger it.
Tune in for the full conversation here
And read below for my three big ideas...
#1: Simplicity Hides ComplexityCopy anchor linkCopied
Modern payment experiences are designed so that the consumer never has to think about the payment itself. The less friction on the front end, the more work the infrastructure on the back end needs to do.
Delivery is a great example. A user opens DoorDash, places an order, the driver picks up the food, pays for it, and, a short while later, the food appears. What the user never sees is the credential created for that transaction and configured so it can only be used in that narrow flow. That credential is what eliminates driver fraud and enforces the limits of the purchase. The complexity sits inside the payment instrument, not the user interface.
The same dynamic appears in BNPL. What looks like a simple checkout button is actually a BNPL provider paying the merchant through issuing rails, removing the need to integrate with each merchant individually.
As front-end experiences become more frictionless, more of the complexity moves into the instrument that carries the payment.
This raises a practical question for agentic commerce: If the issuing infrastructure is the place where a user’s instruction becomes a payment with clear boundaries, who becomes the natural home for agentic AI in practice?
Payments already run on an issuing layer that the consumer never sees. But powering the transaction isn’t the whole story. The most interesting problem is deciding how that transaction should behave, what its limits are, and who is allowed to activate it.
These decisions will become even more critical in the era of agentic commerce.
#2: The Template for How Autonomous Spend Will WorkCopy anchor linkCopied
If you want to understand how AI agents will spend on a user’s behalf, start with the part of payments that already governs delegated spending at scale. During the conversation, Mike used the example of corporate spend management to illustrate a spectrum of trust that AI agents will need to find their place in.
Depending on the level of trust, a credential takes on a different structure: broad permissions when trust is high, tightened controls when trust is partial, and a single-purpose virtual card when trust needs to be narrowly constrained.
This is how companies already allow employees and contractors to spend money while retaining control. The system channels intent, restricts its scope, and supervises its execution.
Agentic commerce maps directly onto this. Rather than sending an agent into the market with a stored card, the safer pattern is to create a virtual card, fund it with a specific amount, set the category and location rules, and bind the agent to that credential. The agent executes the task within those limits. And if something goes wrong, it can initiate a dispute the same way a human would, which lines up with Mike’s point that most disputes now begin inside a chat interface.
This is the context for Marqeta’s Model Context Protocol (MCP) server. It gives an AI agent an accessible way to interact with issuing functions: create a credential, configure its constraints, complete a purchase, or start a dispute.
It’s not a new system of trust. It is a continuation of the one that already governs constrained spend, accounting for hundreds of billions of dollars.
Which raises a more strategic question for the industry. As agents operate on this spectrum, who’s responsible for setting the appropriate level of trust: the user giving the instruction, the issuer provisioning the credential, or the platform coordinating the agent’s behavior?
#3: Customers Want More ControlCopy anchor linkCopied
If you only skim headlines, you might think the consumer economy is flashing warning lights. BNPL use is rising in categories like groceries, and commentators jump straight to the conclusion that households are stretched well beyond their means.
Mike offered a more nuanced reading.
He argued that grocery installments are not a warning sign; it’s a cash-flow strategy. Consumers already do this with credit cards when they pay in full, thirty days later, without revolving. BNPL simply brings that same timing logic into smaller purchases with clearer boundaries. Rising use in non-discretionary categories is not instability. It’s intentional liquidity management.
The second shift reshapes the category entirely. BNPL is moving from a widget on a checkout page to a feature of the credential itself. Affirm and Klarna now issue debit cards that let users elect installments anywhere the network is accepted. Mike noted that this effectively unlocks tens of millions of merchants at once. The installment option stops being something a merchant decides to enable. Instead, it becomes something the user carries with them.
Marqeta’s State of Payments Report picks up a parallel pattern on the business side. Small and medium-sized businesses are starting to see payments as a strategic way to automate and sequence operations. They want fewer manual processes. They want tools that help them line up cash flow with the rhythm of their business. In other words, they’re treating payments the same way consumers are treating BNPL: as choreography. The steps matter, but so does the sequence and the pacing.
Timing control is becoming the real differentiator, and it lives at the intersection of issuing, card features, and user choice.
The industry has long treated rewards, interest rates, and credit lines as the levers that shape competition. But the picture that’s emerging now looks different.
And this raises a larger question for the industry. If timing is the value proposition (not checkout placement), if consumers can elect installments anywhere, how should issuers think about responsible use when the primary behavior they are enabling is control over time rather than on-demand spend?
🎬 DIRECTOR'S COMMENTARY
Mike and I recorded this conversation in the thick of Money20/20 buzz, which feels like the ideal environment for discussing the future of payments if you ask me!
If you want to dig deeper into the trends we touched on, Marqeta’s 2025 State of Payments Report is a good place to start (and several of the questions I asked during our interview came straight from that data!).
WHAT I'M LISTENING TO
#1: Is this the End of the CFPB? (Banking with Interest) 🎧Copy anchor linkCopied
If it feels like there is a division at the CFPB between the people who still work there and are, you know, trying to do their jobs, and Russ Vought, who is staging an elaborate performance art piece on how to destroy the federal bureaucracy slowly, you’re right. That’s exactly what’s happening!
Rob Blackwell, Kate Berry, and Evan Weinberger have the latest.
#2: BNC Squared: Skinny Master Accounts, FDIC Loses to Shareholders, & Travel Debit Cards (Bank Nerd Corner) 🎧Copy anchor linkCopied
Kiah and I talk about Governor Waller’s flirtation with the idea of a skinny Fed Master Account (who is this actually for?!?), a bad beat for the FDIC, and the emergence of reward debit cards from airlines.
You know, your typical Bank Nerd Corner stuff :)
Thanks for the read! Let me know what you thought by replying back to this email.
— Alex
Join 2,444 other finance and fintech leaders in the Fintech Takes Network
The conversation doesn't have to stop here
Keep learning and connecting in the Fintech Takes Network
EVENTS | FEED | LIBRARY | DIRECTORY
