Payments infrastructure has survived the internet revolution, absorbed the mobile shock, and adapted — slowly, imperfectly — to each successive wave of digital disruption. But a new force is taking shape that industry veterans, regulators, and technology strategists are increasingly treating not as an incremental upgrade but as a categorical break from everything that came before. Agentic commerce — the capacity of artificial intelligence systems to move beyond answering questions and begin autonomously executing real-world transactions — is arriving faster than the payments industry appears ready to handle.
The comparison to prior technological inflection points is instructive, and the analogy drawn by veteran fintech analyst Chris Skinner on The Finanser is precise in a way that demands serious attention. When the internet emerged, it transformed static pages into dynamic platforms; businesses that understood only brochureware were suddenly obsolete. When smartphones proliferated, every enterprise of any meaningful scale discovered it needed a native app strategy or risk irrelevance. In both cases, the underlying infrastructure of commerce and payments scrambled to keep pace — and largely succeeded, though at considerable cost and delay. The question now is whether the existing payments stack can adapt with equal resilience when the agent placing the order, authorising the payment, and reconciling the transaction is not a human being at all.
From Passive Tool to Active Participant
The conceptual leap at the heart of agentic commerce is deceptively simple but architecturally explosive. Today's Bank for International Settlements-monitored payments systems — card rails, real-time payment networks, correspondent banking chains — were designed with a human operator at every critical decision node. A cardholder presents credentials. A merchant accepts. An acquirer routes. An issuer approves. Fraud detection flags anomalies for human review. Every link in that chain assumes a person who can be identified, authenticated, held liable, and if necessary, prosecuted.
Agentic AI breaks that assumption entirely. An AI agent acting on behalf of a consumer — booking travel, purchasing software licences, renewing subscriptions, managing a supply chain — does not merely facilitate the human decision; it replaces the human decision at the point of transaction. The agent selects the vendor, negotiates terms where possible, triggers the payment, and logs the outcome, potentially across dozens of micro-transactions per hour. The velocity, volume, and identity complexity this creates for existing payment rails is not a stress test. It is a structural incompatibility.
Infrastructure That Now Belongs to Another Era
Skinner's framing — that existing payments infrastructure "suddenly looks like it belongs to another era" — is worth sitting with. This is not hyperbole from a technology optimist. It is a structural observation about fit-for-purpose design. The rails laid down by Visa, Mastercard, SWIFT, and domestic real-time payment schemes were built to handle human-initiated, human-authenticated, human-disputed transactions. They were not designed to accommodate non-human principals acting with delegated authority across multiple jurisdictions simultaneously.
The implications cascade rapidly. Identity frameworks must evolve to accommodate machine-to-machine authentication at scale. Authorisation models built around static credentials and biometric verification need to account for AI agents operating under delegated scopes that may shift dynamically. Fraud prevention systems that rely on recognising human behavioural patterns will need retraining, or replacement, to identify the anomalies created by agentic actors. Dispute resolution — still a fundamentally human, often manual, process — will require entirely new legal and operational frameworks when one AI agent disputes a charge raised by another AI agent on behalf of competing commercial interests.
Regulatory Terrain Is Equally Unprepared
Regulators face a particularly acute version of this challenge. Frameworks such as the European Banking Authority's payment services directives and the European Central Bank's oversight of systemically important payment systems were constructed around the principle of human accountability. When an AI agent causes a financial loss, current frameworks offer limited clarity on who bears liability — the consumer who deployed the agent, the developer who trained it, the platform that hosted it, or the financial institution that honoured its instructions. That ambiguity is not merely a legal inconvenience; it is a systemic risk.
Neobanks and embedded-finance providers — players such as Revolut and Wise, who have built more API-native, programmable architectures than their legacy counterparts — may find themselves better positioned to integrate agentic commerce flows than traditional institutions. But even these organisations were not designed from inception to handle the non-human principal problem at the core of agentic transactions.
What This Means for the Payments Industry
The shift to agentic commerce is not a future scenario to be modelled in a strategy offsite. It is an active development, unfolding now, with early deployments already testing the boundaries of existing payment authorisation frameworks. What the industry must do — urgently — is distinguish between the challenge of convenience (making payments easier for humans) and the challenge of agency (making payments safe, accountable, and legally coherent when humans are no longer in the loop). The former has driven payments innovation for three decades. The latter is the defining challenge of the next decade. Institutions that conflate the two risk building the digital equivalent of a horse-drawn carriage factory at the precise moment the combustion engine arrives.
Written by the editorial team — independent journalism powered by Codego Press.