The Fourth Phase — Agentic Commerce (1 of 4)
Agentic commerce isn't a faster checkout. It's a different actor at the checkout.
I spent the last stretch researching how AI agents are about to change what it means to pay — and it turned into the longest piece I've written for The Payments Corner. Over the next four posts, I want to pull out the ideas that stayed with me most. Here's the first.
For thirty years, digital commerce optimized one thing above all else: the checkout.
Think about the arc. First we digitized the storefront — the catalog, the cart, the card form. Then we compressed the checkout itself: wallets, one-click, card-on-file, network tokens, each shaving friction off the moment of payment. Then we embedded payments inside marketplaces, apps, and platforms until the act of paying nearly disappeared into the experience around it. Three phases, and through all of them one thing never changed: a person was choosing and paying. We just made it faster and smoother each time.
Agentic commerce breaks that pattern. In delegated purchasing, the person no longer acts at every step. You set an intent — "reorder supplies when we run low," "book the trip under these rules," "keep the team provisioned" — and an AI agent transacts on your behalf. Sometimes moments later. Sometimes for weeks, across many purchases you never individually see.
That's not a faster checkout. It's a different actor at the checkout.
And nearly everything in card infrastructure quietly assumes a human is present at the moment of purchase — there to see the price, consent to the charge, and notice when something looks wrong. Take the human out of that moment and the foundational questions change. Who authorized this transaction? Against what limits? How do you prove consent after the fact, when no one clicked "buy"? What does a dispute even look like when the "cardholder" who made the purchase was software following an instruction from three weeks ago?
This isn't hypothetical positioning. The networks have already named the phase and started building for it — Visa, Mastercard, and American Express each shipped agentic frameworks over the past year, with credentials, scoring, and directories built specifically for transactions an agent initiates. The rails are going down now, deliberately, ahead of the volume.
That's what convinced me this isn't a distant thesis for issuers. It's the next environment their cards are going to operate in — and the institutions that decide early how they want to show up in it will shape the terms for everyone who moves later.
More on that over the next three posts. The full research is on The Payments Corner.
Franco Di Pietro
The Payments Corner
30+ years across payments, fintech, banking, and financial infrastructure. Operator-level perspectives on the systems that move money.
Related Insights
Citi's card-linked offer acquisition bets on cost, not on proven category economics
Citi's acquisition of Kard Financial rests on operational and financial logic—lower rewards spend, faster platform build—but the card-linked offer category itself has never demonstrated standalone profitability. Cardlytics, the only pure-play, has burned $1.15bn cumulatively since its 2018 IPO and generated negative shareholder equity last year.
The PayPal Valuation Paradox
Stripe and Advent walked away, PayPal lost nearly 13% in a day, and the takeover premium disappeared. The failed deal exposed a more important question: how much value is the public market actually assigning to Venmo?
Kraken's card is a bet on controlling the funding decision, not the transaction
Kraken's debit card—with its 2% rewards across multiple asset balances—is not a payments product disguised as one. It is a wallet-stickiness play that inverts traditional card architecture: instead of a card accessing a single account, the platform now decides which of many balances funds each purchase, moving the locus of control from the network to the issuer's financial operating system.