Last weekend, a research outfit most of the payments industry had never heard of wiped billions off the market capitalisation of the world's largest card networks.

Citrini Research published "The 2028 Global Intelligence Crisis," a 7,000-word speculative scenario written from the perspective of June 2028. Within days it had racked up over 6.1 million views on X. By Monday's open, Mastercard, Visa, American Express, and DoorDash had each fallen between 4 and 6 percent. The S&P 500 closed down more than 1 percent.

The thesis was simple, and that is what made it dangerous. When AI agents control commerce, the 2 to 3 percent interchange fee becomes the most obvious cost to eliminate. And stablecoins offer a settlement layer that costs fractions of a penny.

To understand why this scenario rattled the markets, you need to understand how a card payment actually gets paid for.

On a $100 purchase, the merchant typically loses roughly $2.50 in fees. The issuing bank collects the largest share, around 1.7 to 2 percent. The card network (Visa or Mastercard) takes approximately 15 basis points. The acquiring bank or payment processor takes another 3 to 5 basis points. These fees fund the entire card ecosystem, from fraud protection to rewards programmes to the infrastructure that makes tap-to-pay work in 200 countries.

Now consider the alternative. A stablecoin transaction settled on Solana or an Ethereum Layer 2 network costs fractions of a penny. Settlement is near-instant. The rails operate around the clock.

Stablecoins processed $9 trillion in payments in 2025, an 87 percent jump from 2024, according to QED Investors. A separate report from CEX.IO found that stablecoin transfers reached $27.6 trillion in 2024, outpacing Visa and Mastercard's combined transaction volume by 7.68 percent.

Here is the part most coverage gets wrong. The threat is not stablecoins alone. Stablecoins have existed for years without meaningfully denting card network revenues. The threat is what happens when AI agents start making purchasing decisions.

Human shoppers do not optimise for interchange costs. We like our rewards points. We are loyal to card brands. We barely notice the merchant discount rate because we never see it. That friction is invisible to us, and it is precisely what funds the card ecosystem.

AI agents do not think this way. An agent optimising a transaction on your behalf has no loyalty to Visa or Amex. It has no emotional attachment to airline miles. It evaluates cost, speed, and reliability. And when it does, a 2 to 3 percent fee on every transaction looks like waste.

As Citrini wrote, "Once agents controlled the transaction, they went looking for bigger paperclips."

Louis Amira, CEO of Circuit & Chisel, an agentic commerce platform backed by Stripe and Coinbase, told Benzinga the economic logic is sound. In machine-to-machine commerce, a 2 to 3 percent interchange fee is significant, especially when stablecoins can settle near-instantly at a fraction of the cost.

This is not a distant hypothetical. Visa has predicted that millions of consumers will use AI agents to complete purchases by the 2026 holiday season. Nearly half of US shoppers already use AI tools for at least one shopping task. The shift from AI-assisted browsing to AI-executed purchasing is already underway.

The card networks are not standing still. In fact, they are moving faster on agentic commerce than on almost any infrastructure shift in recent memory.

Visa launched its Intelligent Commerce initiative and Trusted Agent Protocol, built on Cloudflare's Web Bot Auth technology. The protocol is already live on Visa's Developer Center and GitHub, with backing from Microsoft, Shopify, Stripe, and Worldpay. Visa is working with more than 100 partners globally, with over 30 actively building in its sandbox and over 20 agent enablers integrating directly.

Mastercard has launched Agent Pay and announced the Mastercard Agent Suite for Q2 2026, combining customisable AI agents with consulting from its global advisory organisation. The company has joined Google's Universal Commerce Protocol and is bringing Agent Pay to Microsoft Copilot Checkout.

Fiserv has integrated Mastercard's Agent Pay Acceptance Framework into its merchant infrastructure, while also partnering with Visa to support its Trusted Agent Protocol. FIS launched an industry-first agentic commerce offering for its issuing bank clients, expected to be available by end of Q1 2026.

The question is whether "trust layer" generates the same economics as "payment rail." Tokenisation, authentication, and fraud controls are valuable. But they may not command 2 to 3 percent of every transaction.

Before we declare the death of interchange, some important nuance.

Visa's own Onchain Analytics Dashboard shows that retail-sized transactions represent less than one percent of all adjusted stablecoin volume. Most stablecoin activity is still high-value transfers, exchange flows, and DeFi activity, not retail commerce.

Crypto card spending has hit $18 billion annualised, growing at a 106 percent compound annual rate since early 2023 according to Artemis Analytics. That is impressive growth, but it runs on existing Visa and Mastercard rails. Users hold stablecoins, but the card network still processes the transaction. Visa captures over 90 percent of on-chain card volume.

As Artemis noted, the real opportunity for stablecoins lies not at the point of sale but in behind-the-scenes settlement. The user experience stays "normal card" while issuers and processors move value in USDC behind the scenes.

BNP Paribas Equity Research reached a similar conclusion: stablecoins do not currently have the potential to disrupt most payment use cases at the checkout. Debit cards already perform well on speed, cost, and access. The disruption potential is strongest in cross-border payments and high-fee categories.

This is where the market reaction to Citrini arguably overshot, and undershot, at the same time.

Visa and Mastercard take only a small number of basis points per transaction. They position themselves as technology companies, not fee collectors. If interchange compresses, the greater pressure falls on the parties collecting the majority of that 2 to 3 percent: the issuing banks and mono-line issuers.

Louis Amira made this distinction explicitly: it is important to separate the networks from the issuers. Most of that fee goes to issuers and other intermediaries. If fees compress, the greater pressure is on those players.

Citrini's scenario flagged American Express, Synchrony Financial, Capital One, and Discover as carrying the sharpest exposure, each falling more than 10 percent in the hypothetical scenario. The reasoning: these companies built entire business segments around rewards programmes funded by the merchant subsidy. If agents route around interchange, those economics collapse.

Mastercard's own response is telling. Days after the Citrini report went viral, the company posted a job listing for a Director of Crypto Flows, tasked with leading stablecoin-linked card issuance, scaling DeFi payment flows, and rewriting network rules for Web3 transactions. CEO Michael Miebach told analysts in January that Mastercard is "leaning in" to stablecoins and agentic commerce, calling the latter a trend where "the train is leaving the station."

The deeper story here is not about any single company. It is about who sets the rules for how AI agents transact.

Right now, we are watching a standards race unfold in real time. Visa's Trusted Agent Protocol, Mastercard's Agent Pay, Google's Agent Payments Protocol (AP2) and Universal Commerce Protocol, OpenAI's Agentic Commerce Protocol, and Coinbase's x402 standard are all competing to define how agents identify themselves, how they authenticate, and how they pay.

Both Visa and Mastercard claim their protocols are complementary, not competitive. Visa says it is aligning with OpenAI's protocol and Coinbase's x402. Mastercard says it is partnering across the value chain, including Google's protocols, to ensure a consistent approach.

But underneath the collaboration language, the stakes are existential. Whoever controls the trust layer for agentic commerce controls the economics of the next era of payments. As commercetools noted, purpose-built agents are already dominating over giant all-in-one agents. Enterprises are adopting small, high-trust agents embedded into existing workflows. The infrastructure that those agents connect to will determine where the fees flow.

The Citrini scenario pointed to Mastercard's Q1 2027 earnings as the potential inflection point, when management might first cite "agent-led price optimisation" and "pressure in discretionary categories." Whether that timeline proves accurate or not, the direction of travel is clear.

Agentic commerce is coming whether the payments industry is ready or not. The question is whether card networks can evolve fast enough to own the trust layer, or whether AI agents will route around them entirely. Where do you think the fees end up?