AI intelligent agents are seizing Visa's market share

Written by: Thejaswini MA

Translated by: Plain Language Blockchain

Visa’s entire business is about betting on behavior. It involves human consumption habits and psychology. The rewards points you collect, the human reward protections you rely on, the Amex Centurion Black Card you aspire to, and the “zero liability” policy that makes you feel safe when withdrawing cash at foreign ATMs — these exist not because “funds transfer” is difficult, but because humans are anxious, status-driven, and poor at reading terms and conditions. Visa has built a $500 billion company within this tower.

However, AI agents lack these traits entirely.

They don’t collect points, aren’t safer because of fraud protections, and don’t have black cards. They have only one command: complete any task. When a task involves payment, the interface performs calculations humans are too lazy to do: the cheapest route, the fastest settlement, the lowest fees. Every time, automatically, emotionless, precise.

Global AI Crisis 2028

Last month, a SubStack article titled “Global AI Crisis 2028” caused Visa’s stock to drop 4%, Mastercard to fall 6%, and American Express to plummet 12%. Although the report claimed this was an “imaginative scenario” rather than a prediction, the market didn’t buy it. The technical statement itself isn’t the point; the core issue is that by 2027, AI agents will bypass the Tokyo system (interchange) and settle using stablecoins. It took Visa fifty years to develop a perfect product for a customer base that is now being replaced.

In “machine-to-machine” (M2M) commerce, a 2-3% card fee (interchange rate) is an extremely visible attack vector. As Citrini Research states: this isn’t to say AI will destroy Visa tomorrow, but that Visa’s fee structure—built on human irrationality—is essentially a tax on human error, while intelligent agents are perfectly rational entities. That’s their significance.

What is Visa selling?

To understand why this is so critical, you must understand what interchange fees actually are. When you buy something with a credit card, the merchant pays 2-3% to the card network and issuing bank. That money funds your rewards points, purchase protections, shopping insurance, and dispute resolution services. The entire consumer value of credit cards is perceived as being paid for by the act of buying; merchants pass the cost onto consumers through slightly higher prices. A beautiful, stable system that has operated for fifty years, where humans in transactions foot the bill—just not directly.

AI agents don’t need these things. They don’t initiate disputes, nor do they require cash-back rewards. The protections that underpin interchange fees are essentially defenses against human errors, fraud, and alarms. Once humans are removed from transactions, these fees are entirely lost in value.

American Express exemplifies this problem most clearly. Its clients are high-income, high-spending, status-seeking elites. Its rates are higher than Visa or Mastercard because its customers are willing to pay for identity and status privileges. The entire model assumes conscious purchasing decisions—choosing Amex over Visa for lounge access, for example. But intelligent agents won’t choose Amex; they seek the cheapest options to accomplish tasks among high-end clientele. In a software-based card world, the so-called hierarchy simply doesn’t exist.

Business bypassing interchange fees driven by intelligent agents poses a huge threat to banks and issuing institutions that rely on related revenue. Much of their profit comes from these 2-3% fees, built around autonomous rewards programs. Visa and Mastercard can develop new network businesses, but those issuing banks whose profit-and-loss models are entirely built on interchange and rewards are out of options.

One-week shipment volume

Citrini’s report, combined with infrastructure launches, coincidentally occurred within a three-week window.

Tempo launched on Wednesday. It’s a payment blockchain developed by Stripe in partnership with Paradigm, designed specifically for high-frequency stablecoin settlement.

Simultaneously, the Machine Payments Protocol was introduced—an open standard allowing AI agents to autonomously pay, consuming human-confirmed steps. It introduces “Sessions”: humans authorize a single spending limit, and the agent continues streaming micro-payments as it consumes data, computations, or API calls. It’s like “OAuth for money”: authorization flows, agent spending, each step consuming a card swipe.

Anthropic, DoorDash, Mastercard, Nubank, OpenAI, Ramp, Revolut, Shopify, Standard Chartered, and Visa have all been default partners of Tempo. The entire payment stack recognizes this structural shift.

On the same day Tempo launched, Visa’s crypto division released a command-line interface (CLI) tool for AI agents. Agents can pay directly from the terminal, without API switches, accounts, or human approval for individual transactions. Visa calls it “Command Line Commerce”—machines transacting without human intervention.

Additionally:

Mastercard agreed to acquire stablecoin infrastructure company BVNK for $1.8 billion.

Circle tested Nano payments online—sub-cent, zero-Gas USDC transactions designed for AI agents to pay per API call.

Sam Altman’s world (originally Worldcoin) launched AgentKit, enabling agents to demonstrate cryptographic proof of representing real humans, integrated directly into Coinbase’s payment infrastructure.

In my view, what happened this week is that all companies are racing to become the new Visa, because Visa has realized it has already lost everything.

The Origin of Obedience

One thing remains unclarified: Visa is not sitting idly.

It’s involved in Tempo’s Machine Payments Protocol, established the Visa Crypto Lab, and its crypto lead explained in Fortune how agents can pay via new standards using synchronized orbit. Mastercard’s $1.8 billion stablecoin investment, Stripe’s acquisitions of Bridge and Privy—all show that legacy organizations understand this shift and are rushing to embed themselves into new infrastructure before the full wave of technology arrives.

Visa’s argument is: it can expand its own orbit to cover AI transactions before building a new one that makes Visa irrelevant.

This isn’t a flawed argument. Stripe’s volume reached $1.9 trillion in 2025 (up 34% YoY). The channel advantage of card networks (distribution power) is hard to replicate. But I must admit, I’m reluctant to voice this openly, because history shows that as soon as you do, new products emerge that make you look foolish.

The flaw in this argument is that Visa’s channel advantage is built on merchant relationships and consumer trust. Merchants accept Visa because consumers hold Visa; consumers hold Visa because merchants accept it. The core of this flywheel is “humans in transactions.” Once agents become the primary buyers in certain key categories, the flywheel halts. Agents have no brand loyalty, no wallets—they operate on instructions and arrangements. Whatever orbit is cheapest and fastest wins their business, and switching costs are clear.

Data versus Narrative Gap

I want to accurately describe our current stage because narratives often run ahead of the data. Despite the ecosystem valuation of around $7 billion for the x402 protocol (referring to a certain agent payment protocol), on-chain data shows that last week, the protocol processed only $28,000 daily, mostly from testing. That’s nowhere near Visa’s daily volume.

However, x402 has already surpassed 50 million transactions. While individual amounts are tiny, the transaction count indicates the infrastructure is being used, and developers are building on top. Merchant acceptance of AI payments is growing. That’s how payment networks are born.

McKinsey predicts that by 2030, AI agents could be involved in $3 trillion to $5 trillion of global consumer commerce. That estimate may be accurate or overly optimistic. But the undeniable fact remains: AI-driven commerce has yet to scale. The businesses building AI agents, the enterprises deploying them as major buyers, and the large-scale transactions that will truly challenge the fee economy are still in development.

Citrini’s report, which caused market disturbance, is based on a series of reliable domino effects. By 2027, the first quarter’s financial reports may not make AI-driven price optimization seem terrifying. Not yet.

The first impact will be in micro-payments within AI infrastructure, in non-consumer sectors. An AI performing research tasks might call hundreds of API data points in a single session, each costing just a few dollars. Over a week, that could generate $40 in revenue. Traditional card networks can’t handle such transactions. The minimum economic model for these transactions doesn’t work, merchant onboarding doesn’t work, fee structures don’t work. This category of business won’t run on Visa’s rails. It requires entirely new systems, which x402, Nano Payments, and Tempo are building.

As for Citrini’s simulated consumer innovations, they will come later. That depends on AI agents handling a significant share of autonomous spending, which in turn depends on whether humans are willing to trust and authorize purchase decisions to AI agents.

Visa is being disrupted by a “better customer”—a customer that, for those who once built Visa’s great non-racial, ethnic benefits, is not a transaction tax but a tax on human rationality. AI agents are perfect rational entities.

How do I know this is important? Because Visa spent $1.8 billion this week to ensure it won’t be excluded from the answer.