AI agents are no longer just recommendation systems and are increasingly becoming active participants in commercial processes. When software can compare providers, select an offer and initiate a transaction, financial institutions need a payment infrastructure that can recognize and control instructions initiated by AI agents.
Traditional payment systems were designed with human presence in mind at the authorization stage. The customer logs in, confirms the transaction or performs an additional authentication step. Agentic commerce changes this model. The software can operate on the basis of authorization previously granted, while the person or company responsible for the transaction is no longer directly involved when it is requested. Financial institutions must therefore verify not only the transaction itself, but also the scope of the authorization under which the agent operates.
The payment infrastructure must support delegation of authority
A payment environment prepared to support AI agents requires the cooperation of several layers:
- Access via APIwhich enables authorized software to retrieve balance and rate information, submit payment instructions, and check transaction status and billing information.
- Identification and authentication of agentsso that the institution can determine which agent acted and on whose behalf.
- Programmable permissionsincluding transaction limits, approved counterparties, currencies, payment methods and thresholds requiring additional approval.
- Anti-fraud, sanctions and compliance checks in real timecarried out before the transfer of funds.
- Automatic transaction reconciliationwhich links each payment to the appropriate authorization, business purpose, fees, and settlement result.
This structure matters because broad account access is not an appropriate model for standalone software.
A purchasing agent for a corporate client may, for example, pay approved suppliers within pre-defined limits when the scope of these powers does not allow for other transfers. The treasury agent can in turn convert selected assets within certain thresholds, and higher value transactions may still require human approval.
These mechanisms should operate at the level of the payment infrastructure, and not only within the agent itself. This allows the financial institution to maintain an independent layer of control in the event of agent malfunction or breach of security.
Routing and settlements are becoming an element of payment architecture
Agentic commerce can generate payment requests via cards, bank transfers, instant payment systems, stablecoins and other digital assets supported on blockchain networks. Financial institutions therefore need routing mechanisms that determine which payment channels are available and allowed.
The choice may depend on cost, settlement speed, liquidity, merchant acceptance, currency and compliance policy. In one situation, real-time bank payment may be appropriate, in another, stablecoin settlement may be appropriate for a cross-border transaction or system-to-system payment that requires 24/7 availability.
AI systems can operate continuously, so the financial infrastructure must also support activity outside traditional banking hours. Real-time authorization and billing can reduce the gap between continuous software activity and the schedules of traditional processing systems.
Security should also be programmable. Financial institutions must be able to determine who authorized the agent, what activities it can perform, its spending limit, which counterparties it can transact with, and when additional approval is required.
Credentials should be instantly revokable so that limits or permissions can be changed immediately if an agent’s role changes or suspicious activity is detected.
Auditability and transaction reconciliation become more important with scale
As the volume of automated transactions increases, financial institutions need a reliable record of every action initiated by an agent.
This entry should indicate:
- agent responsible for the action,
- the basis for the authorization granted,
- payer,
- contractor,
- the payment method used,
- final settlement status.
For banks and fintech companies serving corporate clients, a large number of transactions can quickly increase operational burden if data does not flow in an orderly manner into treasury, accounting, compliance and reporting systems.
Therefore, automatic transaction reconciliation should be a fundamental element of the architecture.
Institutions must also be able to identify transactions initiated by agents. This allows issuers of payment instruments, acquirers and internal controls to apply appropriate rules and retain evidence of how entitlements were established.
Where can blockchain be used?
Blockchain systems can support agent-initiated payments in situations where 24/7 settlements, programmable transfers, and value transfer using stablecoins are important.
They can also support frequent, low-value transactions between systems – for example, payments for data, computing power, or digital services.
Cards and bank payments will continue to play an important role thanks to wide acceptance, developed operational processes and mature protection mechanisms.
For many institutions, a practical solution will be to combine traditional payment channels with blockchain-based infrastructure.
The key strategic decision concerns which elements to develop internally, which to integrate with an external supplier, and which to implement in a hybrid model.
Large banks can independently develop agent identification systems, authorization mechanisms and internal risk policies, while using specialized providers in the areas of custody, liquidity, connectivity to blockchain networks or settlement infrastructure.
Fintech companies, on the other hand, may be more likely to choose a more integrated model, especially when getting the product to market quickly is crucial.
How to prepare for agentic commerce?
For financial institutions, preparing for such a model means treating software as a new source of payment instructions and building systems capable of identifying, authorizing, monitoring, settling and reconciling these instructions.
When assessing infrastructure, institutions should ask specific questions:
- Can the scope of permissions be precisely defined and immediately withdrawn?
- Does the platform support several payment methods?
- Do control mechanisms operate independently of the agent itself?
- Can the infrastructure support large volumes of low-value transactions?
- Are the compliance, settlement and transaction reconciliation processes automated?
- Can new networks or providers be added without having to rebuild the entire architecture?
Agentic commerce will mean that an increasing part of financial activity will be initiated by software. However, the basic obligations of financial institutions remain unchanged.
Banks and fintechs still need to control access, risk, liquidity, compliance and documentation. The difference is that these mechanisms must operate at the speed of software and through interfaces that automated systems can use.
Institutions that prepare the infrastructure to support precisely defined authorizations, flexible routing, continuous settlements and reliable audit trails will be better prepared to operate in this model.
