When tasks involve payments, intelligent interfaces perform calculations people are too lazy to do: the cheapest route, the fastest settlement, the lowest fees.

Written by: Thejaswini MA

Translated by: White Talk Blockchain

Visa’s entire business is about betting on behavior. It involves human consumption habits and psychology. The reward 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 on this tower.

However, AI agents lack these traits.

They don’t collect points, aren’t safer because of fraud protections, and don’t have black cards. They have only one instruction: complete any task. When tasks involve payments, the smart interface performs calculations people are too lazy to do: the cheapest path, the fastest settlement, the lowest fees. Every time, automatically, emotionless, precise.

2028 Global AI Crisis

Last month, a SubStack article titled “2028 Global AI Crisis” 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 important; the core issue is: by 2027, AI agents will bypass the Tokyo system (interchange) and settle using stablecoins. Visa took 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 target. As Citrini Research states: this isn’t to say AI will destroy Visa tomorrow, but that the fee structure Visa built its empire on is essentially a tax on “human irrationality,” and agents are perfectly rational entities. That’s their significance.

What is Visa selling?

To understand why this is crucial, 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 reward points, purchase protections, shopping insurance, and dispute resolution services. The entire consumer value of credit cards is considered to be funded by the purchase itself, with merchants passing 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 cashback. 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 (Amex) exemplifies this problem. Its clients are high-income, high-spending, status-seeking elites. Its rates are higher than Visa or Mastercard because these clients are willing to pay for status privileges. The entire model assumes conscious purchasing—choosing Amex over Visa for lounge access and status. But agents won’t choose Amex; they seek the cheapest options to complete tasks among high-end crowds. In a software-based card world, the so-called hierarchy simply doesn’t exist.

Business bypassing interchange fees driven by agents poses a huge threat to banks and single-issuer card providers that rely on these fees. A large part of their profits come from the 2-3% fees, built around autonomous reward programs. Visa and Mastercard can develop new network businesses, but those issuers whose profit-and-loss models are entirely built on fees and rewards points will have no way out.

Weekly Shipment Volume

Citrini’s report coincided with a period of intense infrastructure launches.

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

Also introduced was the Machine Payments Protocol, an open standard allowing AI agents to autonomously pay with human approval at each step. The protocol introduces “Sessions”: humans authorize a single spending limit, and the agent continues streaming micro-payments as it consumes data, computes, or calls APIs. It’s like “OAuth for money.” Authorization layout, agent spending, each step consuming a card swipe.

Partners like 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 API calls paid per use by agents.

Sam Altman’s world (formerly Worldcoin) project 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.

Origin of Obedience

One thing that has not yet been explicitly pointed out: Visa is not sitting idly.

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

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

The argument isn’t wrong now. Stripe’s volume reached $1.9 trillion in 2025 (up 34% YoY). The channel advantage of card networks (distribution advantage) is hard to replicate. But I must admit I’m reluctant to voice this openly, because history shows that whenever such views are expressed, 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 core buyers in certain key categories, the flywheel halts. Agents have no brand loyalty, no wallets; they operate on instructions and layouts. The most cost-effective and fastest route wins their business, and switching costs are clear.

Data and Narrative Gap

I want to accurately describe our current stage because narratives often run ahead of data. Despite the ecosystem valuation around x402 protocol (referring to a certain agent payment protocol) being about $7 billion, 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 it. Merchant acceptance of agent 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 undisputed fact is: agent-driven commerce has not yet scaled. The businesses building for agents, companies deploying agents as primary buyers, and large-scale transactions that truly challenge fee economics are still in development.

Citrini’s report, which simulated market disruption, is because a series of reliable chain reactions occurred. By 2027’s first quarter earnings, the idea that price optimization driven by agents could become a nightmare isn’t yet real. Not yet.

The first impacts will occur in micro-payments within AI infrastructure and non-consumer sectors. An agent performing research tasks might call hundreds of specialized data APIs in a single session. Each call costs only 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 orbit. It requires something entirely new, and x402, Nano payments, and Tempo are building it.

As for the consumer innovation Citrini simulated, that will come later. It depends on agents handling a significant proportion of autonomous spending, which in turn depends on whether humans are willing to trust and authorize purchase decisions to agents.

Visa is being disrupted by a “better customer”—a customer for those who once made Visa great for non-racial, ethnic interests. The 2-3% fee isn’t a transaction tax; it’s a tax on human rationality. Agents are perfectly rational entities.

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