AI Agents are Coming for Visa's Lunch
Visa built its empire on the human in the loop. The human is leaving.
Visa’s entire business is a bet on human behaviour. It involves human spending, and human psychology. The rewards points you collect, the fraud protection you rely on, the Centurion card you aspire to, the zero-liability policy that makes you feel secure when tapping your card at a foreign ATM. None of it exists because moving money is hard. It exists because humans are anxious, status-driven, and bad at reading terms and conditions. Visa built a $500 billion company on that gap.
An AI agent, however, has none of those qualities.
It does not collect points. It does not feel safer with fraud protection. It does not aspire to a black card. It has one instruction which is to complete the task. And when the task involves paying for something, the agent does the math that no human would ever bother to do. The cheapest rail. The fastest settlement. The lowest fee. Every time, automatically, without sentiment.
Last month, a SubStack post titled”The 2028 Global Intelligence Crisis” crashed Visa 4%, Mastercard 6%, and American Express 12% in a single trading session. The report was labeled a scenario, not a prediction(their words). The market did not care. The technical claim didn’t matter. The problem was that the agents would route around interchanges by 2027 and settle in stablecoins instead. Visa has spent fifty years perfecting a product for a customer that is being replaced.
In machine-to-machine commerce, the 2-3% interchange rate is an obvious target. That line from Citrini Research is the whole thesis. Not that AI will destroy Visa tomorrow. That the fee structure Visa built its empire on was always a tax on human irrationality, and agents are perfectly rational. That is the whole point of them.
What Visa Selling?
To understand why this matters, you have to understand what the interchange fee actually funds.
When you buy something on a credit card, the merchant pays 2-3% to the card network and your issuing bank. This fee pays for your rewards points, your fraud protection, your purchase insurance, and dispute resolution. The entire consumer value proposition of a credit card is funded by merchants, who pass the cost to everyone through slightly higher prices. It is a beautiful, stable system that has run for fifty years because the human being in the transaction was willing to pay for all of it, just not directly.
An AI agent does not need any of these things. It does not dispute charges or want cash back. The protections that justify the fee are protections against human error, human fraud, and human impulsiveness. Remove the human from the transaction, and the fee loses its justification entirely.
American Express is the cleanest version of this problem. Its customers are premium cardholders who are high-earning, high-spending, and aspirational. Its fees are higher than Visa or Mastercard precisely because its customers are willing to pay for the status and the perks. That whole model assumes a human is making the purchase consciously, choosing Amex over Visa because the lounge access is worth it. An agent does not choose Amex. An agent finds the cheapest option that closes the task. The premium tier does not exist in a world where software holds the card.
Agent-driven commerce routing around interchange poses a much greater risk to card-focused banks and mono-line issuers, who rely heavily on the majority of that 2-3% fee and have built entire business segments around rewards programs funded by the merchant subsidy. Visa and Mastercard have network businesses that can adapt. The issuers who built entire profit-and-loss models (P&Ls) around interchange and rewards have nowhere to go.
The week everyone shipped at once
The Citrini report and the infrastructure launches landed in the same three-week window.
Tempo went live on mainnet on Wednesday. Stripe and Paradigm’s payments blockchain, purpose-built for high-volume stablecoin settlement, launched alongside the Machine Payments Protocol which is an open standard that lets AI agents pay for services autonomously, without human sign-off at each step. The protocol introduces sessions. An agent authorises a spending cap once, then streams micropayments continuously as it consumes services like data, compute, or API calls. OAuth for money. The human authorises a budget. The agent spends it. No card required at each step.
Anthropic, DoorDash, Mastercard, Nubank, OpenAI, Ramp, Revolut, Shopify, Standard Chartered, and Visa were all listed as Tempo design partners. The entire payments and commerce stack acknowledging the structural change.
The same day Tempo went live, Visa’s crypto division launched a command-line interface tool for AI agents to make payments from a terminal without API keys, without accounts, without a human authorising each transaction. Visa framed it as “Command Line Commerce” - machines transacting without human intervention.
Mastercard agreed to acquire BVNK, a stablecoin infrastructure startup, for $1.8 billion. Circle launched Nanopayments on testnet. Sub-cent, gas-free USDC transactions built specifically for agents paying for pay-per-call APIs without accounts or credentials. Sam Altman’s World project launched AgentKit, allowing agents to carry cryptographic proof they represent a real human, integrated directly into Coinbase’s payment rails, enabling platforms to verify agent identity without blocking legitimate commerce.
In my view, what happened this week was companies racing to become the new Visa before Visa figures out what it has already lost.
The Paradox in Plain Sight
Now what is not said clearly enough is that Visa is not standing still.
It contributed to Tempo’s Machine Payments Protocol, launched Visa Crypto Labs and its head of crypto was quoted in Fortune explaining how agents can pay with card rails through the new standard. Mastercard is spending $1.8 billion in stablecoin infrastructure. Stripe has acquired Bridge and Privy. The incumbents understand the shift and are positioning themselves in the new infrastructure before it fully arrives.
Visa’s argument is that it can extend its rails to cover agent-driven commerce before agent-driven commerce builds rails that makes Visa irrelevant.
That argument is not obviously wrong. Stripe processed $1.9 trillion in total payment volume in 2025, a 34% year-on-year increase. These companies are not shrinking. The card network’s distribution advantage is not easily replicated. I’ll admit I’m reluctant to say that out loud, because historically, the moment someone makes that argument, something ships and makes them look stupid.
So here is the gap in that argument: Visa’s distribution advantage is built on merchant relationships and consumer trust. Merchants accept Visa because consumers carry Visa. Consumers carry Visa because merchants accept it. The entire flywheel runs on the human being in the loop. The moment agents become the primary buyers in a meaningful category of commerce, the flywheel slows. Agents do not have brand loyalty, nor wallets. What they have are budgets and instructions. Whatever rail is cheapest and fastest wins their business every time, with no switching costs.
I want to be precise about where we are, because the narrative is running ahead of the data.
Despite a roughly $7 billion ecosystem valuation around x402, on-chain data shows the protocol currently processes only about $28,000 in daily volume last week, much of it from testing rather than real commerce. That number, and what Visa processes in a single day, are in completely different leagues.
x402 has crossed 50 million transactions. The volume per transaction is tiny. But the transaction count suggests that the infrastructure is being used. Developers are building on it. The merchant side of the equation (services that accept agent payments) is growing. That is how payment networks start.
McKinsey has estimated AI agents could mediate between $3 trillion and $5 trillion of global consumer commerce by 2030. That estimate may be right. It may be wildly optimistic. What is not in dispute is that agent-driven commerce does not exist at scale today. The merchants building agent-native services, the enterprises deploying agents as primary buyers, the volume that would actually stress-test interchange economics, that is still being built.
The Citrini report spooked the market because it modelled a credible sequence of events. Mastercard’s Q1 2027 earnings are not going to cite “agent-led price optimisation” as the reason volume slowed. Not yet.
The first place this hits is micropayments for AI infrastructure, not consumer commerce.
An agent completing a research task calls specialised data APIs hundreds of times per session. Each call costs fractions of a cent. Over a week, those calls might generate $40 in revenue for the developer running the service. Card networks cannot handle this. The minimum transaction economics do not work. The merchant onboarding process does not work. The fee structure does not work. This category of commerce was never going to run on Visa rails. It needed something new, and x402, Nanopayments, and Tempo are building it.
The disruption to consumer commerce, as Citrini modelled it, comes later, if it comes at all. It requires agents to handle a meaningful share of discretionary spending, which in turn requires consumers to trust agents with purchasing decisions they currently make themselves.
Visa is being disrupted by a better customer. One that has no use for what made Visa great. The 2-3% interchange fee is not a tax on transactions. It is a tax on human irrationality. Agents are perfectly rational.
How do I know this will matter? Because Visa spent $1.8 billion this week making sure it is not left out of the answer.
Token Dispatch is a daily crypto newsletter handpicked and crafted with love by human bots. If you want to reach out to 200,000+ subscriber community of the Token Dispatch, you can explore the partnership opportunities with us 🙌
📩 Fill out this form to submit your details and book a meeting with us directly.
Disclaimer: This newsletter contains analysis and opinions of the author. Content is for informational purposes only, not financial advice. Trading crypto involves substantial risk - your capital is at risk. Do your own research.