There is a version of the agentic commerce conversation that goes like this: autonomous AI agents will begin making purchases on behalf of consumers, the existing dispute infrastructure will not be able to handle it and the industry needs to act before the problem arrives.
That framing is understandable but it’s also slightly wrong, in a way that matters. The problem has not arrived with agentic commerce, as it’s already here. Agentic commerce is simply making it impossible to ignore.
What the current infrastructure was built to do
To understand why, it helps to think about what the current dispute infrastructure was actually designed to do. When a consumer makes a purchase, the transaction generates a trail of evidence: authentication events, session data, click sequences, device identifiers, behavioural signals. That trail exists because a human being was present and active at the moment of purchase. When a dispute arises later, that evidence is what makes arbitration possible. It is how a merchant can demonstrate that the goods were ordered, delivered and received. It is how a bank can assess whether a chargeback claim is legitimate.
The entire architecture of post-transaction dispute management rests on the assumption that intent is legible because a human being expressed it directly, at a specific moment, through a specific set of actions.
Where agentic commerce breaks that assumption
Agentic commerce breaks that assumption cleanly. When an AI agent executes a purchase, the consumer’s intent was expressed at some earlier point (when they granted the agent its permissions and defined its parameters). The agent then acted on those permissions, potentially hours, days or weeks later, without any further human involvement. Whether that action was genuinely authorised is no longer answered by what happened at checkout, it’s answered by what the permission architecture captured beforehand and whether that architecture generated a record legible enough to be used as evidence in a dispute. Most current systems cannot do that.
What merchant data already reveals
Payments infrastructure has become exceptionally good at moving money but it’s still surprisingly poor at explaining why it moved. The data from merchants operating today illustrates this clearly, before agentic commerce has meaningfully scaled. Research from the 2026 Chargeback Field Report, drawing on survey responses from more than 250 merchants, found that almost one in four merchants report using five or more separate systems when compiling evidence for a disputed transaction. Fewer than one in four describe their teams as well-informed on card network rules.
Rather than agentic commerce issues, they are existing infrastructure problems. The evidence is fragmented, the processes are manual and the institutional knowledge is shallow. The systems were not designed to produce a coherent, defensible account of what happened after a payment cleared. And if that is the baseline, the introduction of autonomous purchasing agents applies pressure to a system that was already struggling, at a scale and speed the existing architecture simply cannot absorb.
Authorisation-Authentication-Evidence
The payments industry solved the first two but it’s only now beginning to confront the third.
Where authorisation solved the question of whether a payment could proceed, authentication solved the question of whether the person initiating it was who they claimed to be. But what neither addressed was the question that matters most when a dispute arises: was this transaction genuinely intended, by whom, under what delegated authority and can that be demonstrated from the transaction record itself?
What the payments industry actually needs to build is an evidence layer, or a structured approach to capturing permission, intent and delegation within the transaction record itself, designed from the outset to be legible to dispute processes. That means permission records that capture the scope and limits of any delegated purchasing authority at the moment it is granted. Intent signals that persist across the full transaction lifecycle and connectivity between the systems that currently hold fragments of that evidence in isolation.
Our technology already operates across this layer, connecting fragmented evidence sources, applying AI and machine learning to identify patterns across the full dispute lifecycle, and generating the operational visibility that merchants currently lack. The architecture existsbut the broader payments industry now needs to decide is whether it builds equivalent infrastructure before agentic commerce makes the absence unmistakable.
Solving the right problem before it becomes the loudest one
Agentic commerce makes that priority unavoidable. When the volume of autonomous transactions grows large enough, the inadequacy of current dispute infrastructure will become an industry-wide structural weakness. The time to build the evidence layer is before that point. Overall, the industry has a habit of solving the loudest problem but this time it needs to solve the right one.
