Artificial intelligence agents are about to take market share away from Visa

Article author: Thejaswini M A

Article compiled by: Block unicorn

Introduction

Visa’s entire business model is a bet on human behavior. It’s about human spending and human psychology. The rewards points you accumulate, the fraud protection you rely on, your coveted Centurion card, and the zero-liability policy that makes card payments at ATMs abroad feel reassuring—none of this exists because moving money is difficult. It exists because humans are anxious, seek status, and don’t read the terms and conditions. Visa has built a $50 billion market-cap company by exploiting this gap in cognition.

However, AI agents don’t have these traits.

They don’t accumulate points. They don’t pursue fraud protection. They don’t crave black cards. They have only one instruction: get the job done. And when the job involves payments, the agent performs complex calculations that humans will never bother to run: the cheapest route, the fastest settlement, the lowest fees. Every time. Automated. No emotion.

Last month, a SubStack article titled “The 2028 Global Intelligence Crisis” caused Visa’s stock to drop 4% in a single trading day, Mastercard to fall 6%, and American Express to drop 12%. The report was labeled as “scenario analysis,” not “forecast” (as stated in the original text). But the market isn’t buying it. The technical claims don’t matter. The issue is that by 2027, agents will bypass payment networks and settle using stablecoins. Visa spent fifty years perfecting its product, and now its customer base is being replaced.

In machine-to-machine commerce, a 2–3% interchange rate is obviously a target. That’s the core argument of Citrini Research’s assertion. This doesn’t mean AI will destroy Visa tomorrow. It means Visa’s fee structure—the very foundation on which it built its business empire—is, in essence, a tax on human irrational behavior, while traders themselves are completely rational. That’s what Visa is really about.

What is Visa selling?

To understand why this matters, you need to know what interchange fees are actually used for.

When you shop with a credit card, the merchant pays a 2–3% fee to the credit card network and to your issuing bank. That fee is used to pay for your rewards, fraud protection, purchase insurance, and chargeback dispute resolution services. The entire value proposition to consumers is borne by the merchant, and the merchant ultimately passes the cost on to consumers by slightly increasing the price of goods. It’s a polished, stable system that’s been running for fifty years, because consumers in the transaction are willing to bear all these costs—even if they don’t pay them directly.

AI agents don’t need any of this. They don’t object to fees. They don’t demand refunds. Charging this fee is justified because it can guard against human error, fraud, and impulsive behavior. If no humans are involved in the transaction, the fee loses all meaning.

American Express is the clearest example of this problem. Its customers are high-income, high-spending, status-seeking premium cardholders. Its annual fee is higher than Visa or Mastercard, precisely because its customers are willing to pay for identity and privileges. The premise of this model is that purchases are driven by human motives—customers choose American Express over Visa because access to the Centurion lounges is worth it. Agents won’t proactively choose American Express. They’ll simply look for the cheapest way to complete the transaction. In a world where software controls credit cards, premium membership tiers don’t exist.

A larger risk is posed by credit card banks and single-line issuing institutions that build their entire business around agent-led commercial routing that bypasses interchange fees, rely heavily on 2–3% fee revenue, and structure their business around merchant-subsidized rewards programs. Visa and Mastercard have network businesses that can adapt. But issuing institutions that build their entire profit-and-loss model around interchange and rewards have nowhere to go.

The week when everyone shipped at the same time

The Citrini report and infrastructure projects were published within the same three-week window.

Tempo went live on the mainnet last Wednesday. The payments blockchain developed jointly by Stripe and Paradigm is designed specifically for high-volume stablecoin settlement, and it launched in sync with the Machine Payment Protocol (MPP). MPP is an open standard that allows AI agents to autonomously pay service fees without needing humans to approve each step individually. The protocol introduces session mechanics. Agents only need to authorize a spending cap once, and then they can continue making micropayments as they consume data, perform computation, or make API calls. OAuth authentication is used for payments. The user authorizes a budget, and the agent can spend it. The whole process doesn’t require using a bank card at every step.

Anthropic, DoorDash, Mastercard, Nubank, OpenAI, Ramp, Revolut, Shopify, Standard Chartered Bank, and Visa are all listed as Tempo’s design partners. The entire payments and e-commerce ecosystem has endorsed this structural shift.

On the same day Tempo launched, Visa’s crypto division rolled out a command-line interface tool that lets AI agents make payments via terminals—no API keys, no accounts, and no human authorization required. Visa calls it “command-line commerce”—transactions that machines can conduct without human intervention.

Mastercard agreed to acquire stablecoin infrastructure startup BVNK for $180 million. Circle launched Nanopayments on its testnet—an agent-built, pay-per-use API that can be used without accounts or credentials, priced below a cent and with no Gas fees, for USDC transactions. Sam Altman’s World project launched AgentKit, enabling agents to carry cryptographic proofs to prove they act on behalf of real people; the toolkit is directly integrated into Coinbase’s payments system, allowing the platform to verify an agent’s identity without getting in the way of legitimate transactions.

In my view, what happened last week is that companies raced to become the new Visa so that Visa wouldn’t realize what it has already lost.

The obvious paradox

Now there’s not a single point that isn’t clear enough: Visa isn’t standing still.

It’s been involved in developing the Tempo Machine Payment Protocol (MAPPS), launched Visa Crypto Labs, and its crypto lead has also written an article in Fortune explaining how agents can use bank cards through new standards. Mastercard is investing $1.8 billion in stablecoin infrastructure. Stripe acquired Bridge and Privy. Existing enterprises have already recognized this shift and made preparations before the new infrastructure arrives in full.

Visa’s argument is that it can extend its rails into agent-driven commerce before they become irrelevant—before an agent-led business builds an alternative pathway that makes Visa unnecessary.

This claim isn’t entirely wrong. In 2025, Stripe handled $1.9 trillion in total payments, up 34%. These companies aren’t shrinking. The network distribution advantage of card organizations is hard to replicate. I’ll admit I’m reluctant to say this publicly, because historically, whenever someone makes this argument, new products get released that make them look foolish.

So the flaw in this argument is here: Visa’s distribution advantage is built on relationships with merchants and consumer trust. Merchants accept Visa because consumers hold Visa; consumers hold Visa because merchants accept Visa. The entire loop depends on people. Once agents become the primary buyer in a major commercial category, this flywheel will slow down. Agents don’t have brand loyalty or a wallet. All they have is budgets and instructions. Whichever route is cheapest and fastest will win their business, with zero switching costs.

I want to be precise about where we are, because the pace of public opinion is currently outstripping the data itself.

Even though the ecosystem valuation around x402 is about $7 billion, on-chain data shows that last week the protocol’s daily transaction volume was only around $28,000, most of which came from testing rather than real transactions. Compared with Visa’s daily transaction volume, the gap is enormous.

x402’s transaction volume has already broken 50 million. While the amount per transaction is small, the number of transactions indicates that this infrastructure is being used. Developers are building on top of it. Merchant-side services that accept agent payments are also growing. That’s how payment networks begin.

McKinsey estimates that by 2030, AI agents may facilitate $3 trillion to $5 trillion in global consumer transactions. That estimate may be correct, or it may be overly optimistic. But there’s no denying that agent-driven business models haven’t been widely adopted at scale yet. Merchants building native agent services, enterprises treating agents as their primary buyers, and transaction volumes that can truly validate the economics of transactions are still in development.

The reason Citrini’s report sparked market panic is that it simulated a chain of plausible events. Mastercard’s Q1 2027 earnings won’t blame the slowdown in transaction volume on “agent-led pricing optimization.” At least not yet.

First to be affected is micro-payments infrastructure, not consumer commerce.

Agents completing research tasks call hundreds of specialized data APIs in every session. The cost per API call is only a fraction of a cent. Over a week, these calls could generate $40 in revenue for the developers operating that service. The credit card network can’t handle this. The economics model based on minimum transaction amounts doesn’t work. The merchant onboarding flow doesn’t work. The fee structure doesn’t work. These business models are doomed to fail within Visa’s framework. It needs an entirely new model, and x402, Nanopayments, and Tempo are building it.

As the model Citrini constructed suggests, even if disruption to consumer commerce happens, it will come even later. It requires agents to handle a significant portion of discretionary spending—and that, in turn, requires consumers to trust agents, handing over purchasing decisions they currently make themselves.

Visa is being hit by better customers. These customers no longer need the elements on which Visa’s success depends. The 2–3% interchange fee isn’t a transaction tax—it’s a tax on human irrational behavior, and agents are completely rational.

How do I know this matters? Because last week Visa spent $1.8 billion to ensure it won’t be left out of the answer.