The Agentic Payments Alliance: A Standard to Tax the Machine Economy

The news that Visa, Mastercard, and Fiserv have jointly created the Agentic Payments Alliance is not a technological breakthrough. It is a declaration of war—against a future where autonomous AI agents settle transactions without ever touching a plastic card or a legacy clearing house. For three decades, these companies have functioned as the toll roads of human-initiated commerce. Visa and Mastercard own the card networks; Fiserv sits inside the core processing systems of thousands of financial institutions. Together, they are the plumbing of modern finance. But agentic AI—algorithms that negotiate prices, trigger payments, and manage budgets with zero human oversight—threatens to make that plumbing obsolete. An AI agent does not care about your card metadata. It cares about the cheapest, fastest, and most trustless settlement rail available. That rail may very well be a stablecoin network or a smart contract, not a VisaNet transaction. The alliance's stated goal is to define standards for these machine-originated payments. That sounds benign. But anyone who has audited the history of payment standards knows that standardization is the weapon of incumbents. The EMV chip standard did not become ubiquitous because of technical superiority; it became ubiquitous because a closed licensing group locked out competitors and extracted royalties for decades. The Agentic Payments Alliance is a reincarnation of that playbook, aimed at the most consequential new payment category since the internet itself. The strategic timing is no accident. Regulatory frameworks for AI are still being written—the EU AI Act is implementing in phases, China is refining its generative AI rules, and the U.S. is a patchwork of state statutes. In this vacuum, the incumbents are not waiting for regulators. They are writing the rules themselves. This is the financial equivalent of the private sector drafting its own jurisprudence before governments even realize the trial has started. Let me be precise about the technical fault lines. The current payment infrastructure—ISO 8583 for card transactions, ISO 20022 for modern messages, the global real-time settlement systems—was built for human-in-the-loop decision making. The entire risk architecture assumes a human pattern: a user logs in, authenticates, approves a transaction. AI agents break that assumption. They operate with human-out-of-the-loop autonomy, which means the well-understood fraud signals—device fingerprints, typing cadence, behavioral biometrics—become meaningless. The alliance will need to build an identity trust model for machines, not people. That is not a trivial extension of existing systems; it is a fundamental redesign. And here is the part that keeps me up at night: AI agents are designed to optimize one objective. When thousands of them share the same training data or the same model provider, they will inevitably exhibit herd behavior. A single model's misjudgment could trigger synchronized buying, selling, or settlement across millions of agents simultaneously, creating a liquidity pulse that no existing clearing system is designed to absorb. The alliance's initial framework—assuming it ever sees the light of day—will likely focus on the mundane plumbing of microtransactions. But the systemic risk lies precisely in the collective action problem, not the individual transaction. The business logic of the alliance is even more problematic. Visa and Mastercard charge fees based on transaction value. But machine-to-machine payments are characterized by an exponential increase in volume and a collapse in per-transaction value. A supply chain AI might execute 10,000 micro-settlements a day, each worth fractions of a cent. The traditional percentage-based toll model collapses under that weight. The alliance will have to invent a new fee structure—perhaps a hybrid model that includes per-transaction, per-value, and per-intelligence fees. That is a polite way of saying they will find a way to tax the intelligence of the agent itself. Meanwhile, Fiserv's presence is the strategic sleight of hand. Visa and Mastercard own the major arteries of card commerce, but Fiserv owns the capillaries—the core processing software that runs inside tens of thousands of small banks. By embedding the alliance's standards into those banking backends, the incumbents ensure that even if an AI agent tries to bypass the card networks, it will still hit Fiserv's API. The alliance is not building a new railroad; it is quietly laying switches on every existing track. Now for the contrarian angle. I have spent my career watching incumbents defend moats with standards, but agentic AI is a different beast. The very qualities that make AI agents useful—autonomy, optimization, and refusal to pay unnecessary tolls—will also make them resistant to top-down standardization. The alliance's members are all exquisitely regulated organizations. They will need to build KYC and AML compliance into their AI payment protocols. That means the AI agent must be identified, monitored, and sanctioned at every step. But an AI agent's native environment is a permissionless blockchain, where an agent-controlled wallet can transact directly with another agent's smart contract, with no KYC, no chargebacks, and no intermediary. The alloy of regulatory compliance will create friction. And in the agentic economy, friction is death. The very standards the alliance is designing to capture machine payments will likely drive the most sophisticated AI agents toward decentralized rails—stablecoins, lightning networks, or even entirely new consensus mechanisms for machine-to-machine value transfer. The alliance's standards will become the 'safe' route for legacy enterprises, but the frontier AI agents will go around them. This is not a decoupling thesis; it is a displacement thesis. The card duopoly won the consumer era because they had the best network effects among humans. They cannot win the machine era because machines will choose the cheapest code path, and that path is likely a smart contract. I also see a hidden regulatory opportunity in this alliance. The incumbents are so focused on defending their toll booths that they are missing the bigger game: the definition of liability. When an AI agent authorizes a transaction that goes wrong, who is at fault? The user? The agent's developer? The payment network? This is a legal void. The alliance could have positioned itself as the arbiter of that liability framework, turning a technological threat into a jurisprudential asset. But by focusing on standards for transaction routing, they are ceding the battlefield of liability to the AI model providers and the decentralized protocols. In the long run, whoever defines liability will define the value chain. Visa and Mastercard once prospered because they solved the trust problem for anonymous consumers. In the machine economy, they are failing to solve the trust problem for autonomous software agents, and that failure will be someone else's opportunity. We have been here before. In 2017, I watched projects raise billions of dollars on the back of non-existent smart contracts, while the industry cheered on vapor. The regulators eventually came, and the bubble burst. 2017's dream is today's regulation. The same cycle is now playing out in the AI agent payment space, but with a twist: this time, the incumbents are not on the side of the innovators. They are on the side of the status quo, desperately trying to fit a quantum leap into a classical box. The question is not whether their agents will adopt the alliance's standards. The question is how many generations of AI agents will be trained to treat those standards as an unnecessary tax. The answer, I suspect, is none. Because the first thing any self-respecting agent learns is to minimize costs, and the cost of legacy payment rails is no longer acceptable in the machine economy. The future of AI payments does not need a permission slip from Visa. It will simply write its own. So let's stop celebrating the alliance as a step forward. It is a step backward, a rearguard action to preserve a revenue model that is structurally incompatible with the autonomous future. The true tell is not what they say, but who is missing: no OpenAI, no Google, no Anthropic. The companies actually building the agents are not in the room. That is not an oversight. It is a confession. The alliance is run by toll collectors, not by the architects of the machine economy. The toll collectors are laying new booths, but the road is being built elsewhere. And that road leads to cryptographic trust, not to card scheme member agreements. We are about to find out whether trust can be standardized on a PDF—or whether it has to be written in code.