AI agents are抢夺 Visa’s lunch
TechFlow Selected TechFlow Selected
AI agents are抢夺 Visa’s lunch
When a task involves payment, the intelligent interface performs calculations that people are too lazy to do: identifying the cheapest route, the fastest settlement, and the lowest fees.
Navigating Web3 tides with focused insightsBy: Thejaswini MA
Translated by: Baihua Blockchain
Visa’s entire business is a bet on human behavior—on consumption habits and psychology. The reward points you collect, the fraud protection you rely on, the aspirational American Express Centurion Black Card, and the “zero liability” policy that makes you feel safe withdrawing cash from an ATM overseas—none of these exist because moving money is technically difficult. They exist because humans are anxious, status-driven, and terrible at reading terms and conditions. Visa built a $500 billion company atop this tower of human frailty.
AI agents, however, possess none of these traits.
They don’t collect points. They don’t gain security from fraud protection. They don’t covet black cards. They have one instruction: complete the task. When the task involves payment, intelligent interfaces perform calculations humans are too lazy—or too impatient—to run: cheapest route, fastest settlement, lowest fees. Every time, automatically, emotionlessly, flawlessly.
The 2028 Global Intelligence Crisis
Last month, a SubStack article titled “The 2028 Global Intelligence Crisis” triggered a 4% drop in Visa’s stock price, a 6% decline for Mastercard, and a staggering 12% plunge for American Express. Though the report explicitly framed its scenario as “imaginative fiction,” not prediction, markets weren’t buying it. The technical details themselves mattered less than the core implication: by 2027, AI agents will bypass legacy clearing systems (interchange networks) and settle directly in stablecoins. Visa spent fifty years perfecting a product for a customer base now being replaced.
In machine-to-machine (M2M) commerce, the 2–3% interchange fee is an exceptionally conspicuous attack surface. As Citrini Research put it: “It’s not that AI will destroy Visa tomorrow—but Visa’s empire was built on a fee structure that taxes ‘human irrationality,’ while agents are perfectly rational. That’s precisely their point.”
What Is Visa Selling?
To grasp why this matters, you must first understand what interchange fees really are. When you buy something with a credit card, the merchant pays 2–3% to the card network and issuing bank. That money previously funded your rewards points, purchase protection, shopping insurance, and dispute resolution services. The entire consumer value proposition of credit cards has long been subsidized by purchases—and merchants recoup those costs by slightly raising prices. It’s a beautiful, stable system that’s run for fifty years, with humans paying for everything in the transaction—not directly, but willingly.
AI agents need none of this. They won’t file disputes or demand cashback. The protections underpinning interchange fees are fundamentally defenses against human error, human fraud, and human alerts. Remove humans from the transaction loop, and those fees instantly lose all value.
American Express is the starkest embodiment of this problem. Its customers are high-income, high-spending elites driven by status. Its fees exceed Visa’s or Mastercard’s precisely because those customers willingly pay premiums for prestige and privilege. The whole model assumes conscious, identity-driven purchasing—choosing Amex over Visa for lounge access. Agents won’t choose Amex. Agents scan the high-end cohort for the cheapest option capable of completing the task. In a world of software-issued cards, tiers simply don’t exist.
Agent-driven commerce bypassing interchange fees poses an existential threat to banks and single-product issuers dependent on those fees. A large portion of their profits comes from that 2–3%, and entire business units have been built around proprietary rewards programs. Visa and Mastercard can pivot toward network-based businesses; issuers whose profit-and-loss models hinge entirely on interchange revenue and points economics have nowhere to go.
A Week of Cumulative Shipments
Citrini’s report coincided almost exactly with a burst of infrastructure launches—a three-week window of unprecedented density.
Tempo launched on mainnet this past Wednesday. A payment blockchain co-developed by Stripe and Paradigm, it’s purpose-built for high-frequency stablecoin settlement.
Also launched was the Machine Payments Protocol—an open standard enabling AI agents to make autonomous payments without requiring human confirmation at every step. The protocol introduces “Sessions”: humans grant a one-time spending cap, after which agents stream micro-payments continuously as they consume data, compute, or API calls. Think of it as “OAuth for money.” Humans authorize once; agents spend; each action consumes a fraction of the allowance.
Anthropic, DoorDash, Mastercard, Nubank, OpenAI, Ramp, Revolut, Shopify, Standard Chartered, and Visa were all default partners on Tempo. The entire payments stack acknowledges this structural shift.
On the same day Tempo launched, Visa’s crypto division released a command-line interface (CLI) tool designed specifically for AI agents. Agents can now pay directly from the terminal—no API gateways, no accounts, no per-transaction human authorization. Visa calls it “Command Line Commerce”—transactions executed by machines, without human intervention.
Also:
Mastercard agreed to acquire stablecoin infrastructure firm BVNK for $1.8 billion.
Circle launched Nano Payments in beta—a sub-cent, zero-gas USDC transaction layer designed expressly for AI agents paying per API call.
Sam Altman’s Worldcoin project rolled out AgentKit, allowing agents to cryptographically prove representation of real humans—and integrate directly into Coinbase’s payment rails.
To me, what happened this week is clear: every company is racing to become the new Visa—because Visa realizes it’s already lost everything.
The Obedience Thesis of Origins
One thing remains unspoken: Visa is not sitting still.
It participated in Tempo’s Machine Payments Protocol, founded the Visa Crypto Lab, and its head of crypto even explained in Fortune how agents will use new standards to pay via synchronized rails. Mastercard’s $1.8 billion bet on stablecoin infrastructure, Stripe’s acquisitions of Bridge and Privy—all legacy institutions recognize the shift and are embedding themselves into the new infrastructure before tech fully arrives.
Visa’s thesis is simple: it can expand its existing rails to cover agent transactions before agents build new rails where Visa connectivity becomes irrelevant.
This thesis isn’t wrong—at least not yet. Stripe’s 2025 volume hit $1.9 trillion (up 34% YoY). Card networks’ distribution advantages are hard to replicate. Yet I hesitate to say this aloud—history shows that whenever someone declares something “not wrong yet,” a new product appears and makes them look foolish.
The flaw lies in Visa’s distribution advantage itself: it rests on merchant relationships and consumer trust. Merchants accept Visa because consumers hold Visa; consumers hold Visa because merchants accept it. The flywheel spins only because of “humans-in-the-loop.” Once agents become the primary buyers in key commercial categories, the flywheel stops. Agents have no brand loyalty, no wallets—they have budgets and instructions. Whichever rail is cheapest and fastest wins their business, and switching costs are negligible.
The Gap Between Data and Narrative
I want to accurately describe our current stage—because narratives often sprint far ahead of data. Though the ecosystem around the x402 protocol (note: the original text refers to an agent payment protocol) carries an estimated $7 billion valuation, on-chain data shows the protocol processed just $28,000 daily last week—mostly test traffic. That figure is orders of magnitude smaller than Visa’s daily volume.
Yet x402 has already surpassed 50 million transactions. While individual amounts are trivial, transaction count signals active usage—and developers building on top. Merchant-side adoption of agent payments is growing. This is how payment networks are born.
McKinsey forecasts AI agents could mediate $3–5 trillion in global consumer commerce by 2030. That estimate may be accurate—or overly optimistic. What’s indisputable is this: agent-driven commerce has not yet scaled. Businesses actually serving agents, enterprises deploying agents as primary buyers, and large-scale transactions that truly threaten the interchange economy—are still under construction.
Citrini’s report rattled markets because it triggered a cascade of credible, real-world follow-ons. Q1 2027 earnings reports likely won’t yet reveal price optimization driven by agents as a material risk. It’s not time yet.
The first shock will land in micro-payments for AI infrastructure—non-consumer, non-retail domains. An agent performing research might call hundreds of specialized data APIs in a single session—each costing mere cents. Weekly revenue per agent could reach $40. Traditional card networks cannot process such transactions. Their minimum economic thresholds fail, onboarding processes fail, and fee structures fail. This category of commerce won’t run on Visa’s rails. It demands something entirely new—and x402, Nano Payments, and Tempo are building it.
As for Citrini’s imagined disruption to consumer spending? That arrives later. It requires agents to handle a meaningful share of autonomous spending—which hinges entirely on whether humans are willing to trust and delegate purchasing decisions to agents.
Visa is being disrupted by a “better customer”—one that renders obsolete the very human irrationalities that made Visa great. The 2–3% interchange fee isn’t a transaction tax. It’s a tax on human irrationality. Agents are perfectly rational.
How do I know this matters? Because Visa spent $1.8 billion this week—just to ensure it isn’t excluded from the answer.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
