Existing payment rails were designed around a specific set of assumptions about how transactions happen: a person opens a checkout, decides to buy, enters or confirms payment credentials, and submits. Authentication systems challenge the credentials presented on behalf of an identity. Liability frameworks assign responsibility based on that person’s demonstrated intent. Audit trails record what that person did.
For now, those assumptions mostly hold. Recent NeilsenIQ data shows that 42% of consumers have used at least one AI tool to shop in the past month. But the dominant behavior is still assistive rather than autonomous: AI helping people compare products, surface discounts, and narrow choices before a human makes the final call. Only 5% of consumers report fully autonomous purchasing, where an agent places the order without the consumer present at checkout.
That 5% is where the payment infrastructure assumptions start to break down, and it is also where adoption is heading—there won’t always be a person involved.
Agentic commerce, the use of AI autonomous transactions that browse, select, and complete purchases without the user present, is advancing faster than the payment infrastructure beneath it. Adobe Analytics reports a 4,700% increase in generative AI traffic to US retail sites grew year-over-year in July 2025, and McKinsey projects agentic commerce could orchestrate $3 trillion to $5 trillion globally by 2030.
The current low adoption rate creates a window for software providers (ISVs) with embedded payments to evaluate their infrastructure before that pattern becomes a standard part of the transaction mix, rather than after merchant complaints and unresolved disputes make the gaps visible.
Same rails, different assumptions
On the surface, an agentic transaction moves through the same process as any other: authorization, settlement, reconciliation. The card networks, the acquiring banks, and the payment processors handle it using the same technical sequence they have always used.
But the complications are in the assumptions baked into that sequence:
- Authentication systems are built to challenge a human, such as a passcode or a one-time code sent to a phone. When an agent initiates a transaction, there is no human to receive or respond to those challenges.
- Liability frameworks for payment disputes typically hinge on cardholder intent, the idea that a specific person made an active decision to authorize a charge. Agent-initiated payments introduce ambiguity into that determination that current frameworks were built to resolve only in human-initiated contexts.
- Audit trails, too, are constructed around a human actor, recording IP addresses, device fingerprints, and behavioral signals that map poorly onto an automated process.
Chargebacks and disputes can also become operationally opaque. For example, when a consumer reviews a charge that an AI agent placed on their behalf and fails to recognize it, or disputes whether the agent acted within the parameters they set, the dispute workflow that follows was designed for a different scenario entirely. The cardholder has a reasonable grievance. The merchant has a completed transaction with no human interaction record to reference. The ISV sits in the middle, handling a dispute type that the existing evidence-gathering process was built around human authorization.
Until regulatory frameworks specify how agent authorization is documented and verified, the burden of that ambiguity falls on whoever built the payment environment.
Loading the infrastructure with new responsibility
When humans are present in transactions, they absorb a significant amount of operational friction. A person waits while authorization completes, retries when a payment fails, calls support when something goes wrong, and makes a judgment call when edge cases arise. An agent records a failure and moves on, retrying only when specifically programmed to do so.
For ISVs, the visibility problem builds up on the reliability problem.
When a human-initiated payment fails, the person knows immediately and has context. When an agent-initiated payment fails, the merchant often sees a string of failed transactions without a clear explanation of what went wrong or why. The ISV owns that conversation, which means the payment infrastructure for AI-initiated flows needs to generate the kind of structured, interpretable failure data that a human support interaction would have captured naturally.
Four questions every ISV should ask their payment partner
When agentic commerce surfaces a gap in the stack, the merchant often calls the ISV. That positioning is an advantage when the infrastructure is sound and a liability when it falls short. Make sure you’re prepared to answer the call by assessing your readiness for transactions without a human in the loop:
- Authentication. How does the system handle verification when no cardholder is present to respond to a challenge?
- Dispute workflows. Can the evidence-gathering process accommodate agent-authorization documentation rather than human interaction records?
- Failure visibility. How are failed transactions logged and surfaced to merchants in a way that supports diagnosis and resolution?
- Infrastructure commitments. Do uptime and latency guarantees match the performance requirements of automated, sequential transaction chains?
Dispute handling deserves particular attention. The chargeback exposure created by agent-initiated transactions is already present in early agentic deployments, and it will grow as adoption accelerates. The ISVs best positioned to manage it are those whose payment environments generate clear authorization records, structured audit trails, and dispute workflows capable of presenting agent-specific evidence.
Soon, the regulatory dimension will make it clear what the best approach is. Frameworks governing agentic transactions are still forming, and the eventual requirements will almost certainly address how agent authorization is established and documented before a transaction is submitted. ISVs building on infrastructure that already handles compliance without human checkpoints—through tokenization, hosted payment pages, and structured authentication flows—will have less to retrofit when those requirements arrive.
Conclusion: Infrastructure evaluation as a competitive posture
Agentic commerce reveals the degree to which payment infrastructure has relied on human participation to function correctly. The person at checkout was doing more work than the system ever had to account for: absorbing friction, resolving ambiguity, providing context for disputes, and serving as the reference point for liability determinations. Removing that person exposes the assumptions the system was built on.
At 5% autonomous adoption, that exposure is still manageable. Most agentic activity today is AI assisting a human decision rather than replacing it, which means the edge cases are arriving in small numbers before they arrive at scale. That is the window. ISVs that examine their infrastructure now, before agentic transaction volume makes the gaps operationally visible, will have less to retrofit and more to offer when autonomous purchasing moves from an emerging behavior to a routine one. The payment stack built to handle human-initiated transactions will need to earn that position again under different conditions.
"*" indicates required fields