Beyond Agentic Commerce: The Next Evolution of Payments

By Michelle Brahmbhatt, Founder, Kuar Consultancy

From intelligent payment initiation to autonomous operations, agentic AI is changing how we think about the payment journey.

Agentic commerce is already beginning to change how consumers discover and purchase products. Initiatives such as OpenAI and Stripe's Agentic Commerce Protocol, Google's Agent Payments Protocol, Visa Intelligent Commerce, and Mastercard Agent Pay are helping establish the infrastructure for AI agents to participate in transactions on behalf of consumers.

But purchasing products is only one part of how money moves.

What happens when AI agents begin managing loan payments, paying medical bills, settling invoices, transferring funds between accounts, or coordinating payment exceptions?

And what happens when payment providers use agents internally to investigate authorization declines, manage ACH returns, reconcile settlements, or optimize payment operations?

I see two parallel transformations taking shape: agentic payments, changing how consumers and businesses initiate and manage transactions, and agentic payment operations, changing how payment providers manage the payment lifecycle.

Both present opportunities to rethink existing payment experiences. But realizing their potential requires more than integrating AI into payment workflows.

Beyond the Shopping Cart

Much of the current conversation around agentic payments focuses on commerce: a consumer instructs an AI assistant to find a product, compare options, and complete a purchase within defined parameters.

But consider a different scenario.

A consumer tells an AI assistant:

"Make sure my car loan is paid before the due date each month, but maintain at least $2,000 in my checking account."

Or:

"Pay my outstanding medical invoices under $200 once the final patient balance has been confirmed."

Or even:

"Move excess funds from checking to savings, but account for upcoming bills and expected deposits."

These aren't traditional shopping transactions. They involve financial obligations, money management, payment timing, available funds, authorization, and decision-making.

Agentic capabilities could allow payment experiences to move beyond executing predefined instructions toward evaluating changing conditions and coordinating actions within established permissions.

The opportunity extends to loan servicing, healthcare payments, bill pay, account-to-account transfers, business invoicing, subscription management, and disbursements.

But this raises an important question: What makes these experiences truly agentic rather than another form of payment automation?

From Automation to Autonomous Decision-Making

The payments industry is no stranger to automation.

Recurring payments, automated retries, account updater services, and rules-based payment routing have existed for years.

A recurring payment system can initiate a transaction on a predetermined schedule. An agentic system goes further, evaluating changing conditions, considering eligible payment methods, and coordinating actions toward a defined outcome.

For example, rather than simply initiating a loan payment on the 15th, an agent could evaluate the payment due date, available funds, eligible payment methods, processing timelines, transaction costs, and the customer's established preferences.

The agent could also evaluate the most cost-effective payment rail at the time of the transaction, balancing the cost to the merchant or consumer against settlement speed, transaction risk, and payment acceptance requirements.

This introduces another interesting question: Whose interests should the agent optimize for: the consumer, the merchant, or the payment provider?

A merchant may prefer ACH because of lower processing costs, while a consumer may prefer a credit card for rewards, purchase protections, or cash-flow flexibility. A payment provider may have its own commercial considerations.

Those interests don't always align.

Agentic payment products will need clearly defined objectives, transparent decision-making, and appropriate controls to ensure that payment selection reflects the permissions and interests the agent is intended to serve.

If the original payment plan cannot be completed, the agent might propose an alternative payment method or request additional authorization.

But autonomy must have boundaries.

An agent's ability to determine that a payment should occur does not automatically mean it has the authority to initiate that transaction.

And not every payment workflow needs an AI agent. If conventional automation can reliably achieve the same outcome, introducing agentic AI may add unnecessary complexity and risk.

The product opportunity lies in identifying where contextual reasoning and coordinated decision-making create meaningful value beyond existing automation.

Who Owns the Agent?

As agentic payments expand beyond commerce, another question emerges: where does the consumer interact with the agent, and who provides the underlying capabilities?

A consumer might manage loan payments through an AI assistant embedded in a lender's application. A healthcare patient could use an assistant within a patient portal to manage outstanding balances. A banking customer might instruct an agent to coordinate transfers and bill payments across connected accounts.

Alternatively, a consumer could use a personal AI assistant to manage financial obligations across multiple providers.

These models create different opportunities for banks, lenders, billers, payment processors, ISVs, and PayFacs.

For payment service providers, the strategic question isn't necessarily whether to build an independent AI assistant.

It may be whether to integrate with emerging agentic platforms, expose payment capabilities through APIs, or develop specialized agentic workflows that complement their existing products.

Vertical payment providers may have a particular opportunity here.

A healthcare payments platform, for example, may already operate within workflows involving patient billing, insurance processing, payment plans, refunds, and reconciliation. An agent that understands these workflows could offer more value than one focused solely on initiating a transaction.

The differentiation may not be in owning the agent itself, but in understanding and enabling the payment workflows the agent must navigate.

The Second Transformation: Agentic Payment Operations

While agentic commerce focuses on how customers interact with payments, another transformation is emerging within payment providers themselves.

Consider the operational activities that occur throughout the payment lifecycle.

Authorization declines must be investigated. ACH returns need to be classified and resolved. Settlement discrepancies require reconciliation. Disputes involve gathering evidence, reviewing transaction histories, and coordinating responses.

Many of these processes already benefit from automation, but they often involve multiple systems, fragmented information, and manual decision-making.

Agentic AI could help bridge those gaps.

An agent could investigate declining authorization rates, identify patterns across transactions, and recommend corrective actions.

Another could analyze ACH returns, identify common failure reasons, and coordinate permitted recovery workflows.

A reconciliation agent could investigate unmatched settlement records, trace discrepancies across processor reports and internal ledgers, and escalate unresolved exceptions.

The opportunity isn't simply automating individual tasks. It's enabling coordinated workflows across the payment lifecycle.

But there is a distinction between allowing an agent to investigate and recommend an action and allowing it to independently execute a financially consequential decision.

Changes to routing configurations, payment retries, refunds, fund movements, and other high-impact actions require appropriate authorization, controls, and oversight.

Agentic payment operations should be designed around controlled autonomy, not unrestricted autonomy.

AI Capabilities Must Work Within the Realities of Payment Infrastructure

This is where I believe deep payments product expertise becomes especially important.

An AI agent may determine when to initiate a transaction, which payment method to use, or how to respond to a failed payment.

But payment execution involves more than invoking an API.

ACH, card networks, RTP, and FedNow operate under different rules, processing requirements, settlement models, and risk considerations.

Consider an agent managing a loan payment through ACH.

It must account for processing windows, bank holidays, settlement timing, return risks, and the lender's payment-posting requirements.

More importantly, the transaction must have valid authorization under applicable Nacha rules.

An instruction given to an AI assistant is not automatically sufficient authorization for an ACH debit. The authorization framework, transaction classification, customer consent, record retention, and revocation requirements must all be considered.

The same principle applies to other payment rails.

A card transaction may involve network tokenization, credential permissions, authentication, and dispute rights. Real-time payment rails introduce different considerations around payment initiation, availability, finality, and fraud controls.

And the payment lifecycle doesn't end with successful initiation.

Payments can fail, be returned, require refunds, become disputed, or create reconciliation exceptions. Agentic capabilities must account for those downstream events rather than focusing exclusively on transaction execution.

The challenge isn't simply building agents capable of initiating payments. It's designing agents that understand when and how a payment can be initiated, the rules governing the underlying payment rails, and the operational responsibilities that follow.

This requires product leaders who understand payment rails, regulatory and network requirements, operational dependencies, and the practical complexities of integrating new capabilities into existing infrastructure.

AI expertise and payments domain expertise must work together.

From Agentic Strategy to Product Execution

For payment providers, the emergence of agentic capabilities creates both opportunity and pressure to innovate.

But the question shouldn't be how quickly an organization can introduce AI agents.

It should be where agentic capabilities can deliver measurable value, which problems they are best suited to solve, and how those capabilities can be delivered responsibly.

Product leaders need to evaluate opportunities across customer experience, payment performance, operational efficiency, risk reduction, and potential new revenue streams.

They must also determine what to build internally, where to partner, and which existing platforms or protocols can accelerate delivery.

Not every provider needs to build its own agentic commerce infrastructure. For some, integrating with an established platform may be the appropriate strategy. Others may find greater differentiation in vertical-specific payment workflows or agentic operational capabilities.

Execution brings another level of complexity.

Introducing an agentic payment capability may require changes across customer authentication, payment orchestration, processor integrations, compliance controls, transaction monitoring, reconciliation, and customer support.

It also requires coordination across engineering, risk, compliance, operations, financial institutions, networks, and technology partners.

A sensible approach is to introduce autonomy incrementally: begin with analysis and recommendations, validate outcomes, introduce approved execution, and expand autonomous capabilities as confidence and controls mature.

The objective isn't to maximize autonomy. It's to deliver reliable outcomes while maintaining trust and accountability.

Building an AI Culture Matters as Much as Building AI Capabilities

There's another aspect of this transformation that I believe deserves equal consideration: the culture organizations create around AI.

Companies can invest heavily in AI-native payment products and internally developed agents, but technology investment alone doesn't guarantee meaningful adoption.

AI transformation cannot simply be mandated from the top down.

Requiring teams to use newly introduced agents doesn't necessarily mean those agents solve real problems, improve productivity, or earn the trust of the people expected to use them.

Organizations need a culture of experimentation, where employees are encouraged to explore AI, challenge existing workflows, and identify opportunities within their own areas of expertise.

The people closest to the work often understand its complexities best.

A payments operations analyst investigating ACH returns every day may recognize opportunities that aren't immediately apparent to an AI development team. A risk analyst may identify controls that need to exist before an agent can execute a particular action.

Bringing these teams into discovery, experimentation, and development helps ensure that AI capabilities address meaningful problems rather than simply introducing another tool.

This doesn't mean experimentation should happen without governance. Particularly in payments, organizations need clear boundaries around data access, approved tools, financial decision-making, and human oversight.

But governance and mandatory adoption aren't the same thing.

The goal shouldn't be to make everyone use AI. It should be to create an environment where people understand its potential, feel empowered to experiment, and help shape how it is applied.

Successful experiments can then be validated, standardized, and scaled.

The Road Ahead

Agentic commerce is an important step in the evolution of payments, but I believe it represents only part of a much broader opportunity.

This evolution is already underway. Agentic capabilities are beginning to reshape how consumers and businesses interact with payments, while payment providers are exploring how AI agents can improve transaction processing, operational efficiency, and exception management.

As these capabilities mature, their applications could extend further into financial obligations, account-to-account transfers, invoicing, and the broader payment lifecycle.

Two transformations are taking shape: making payment experiences more intelligent and making payment operations more autonomous.

Both will require thoughtful product strategy, deep payments domain expertise, disciplined execution, and an organizational culture that encourages experimentation.

The underlying payment rails, regulatory frameworks, and operational responsibilities don't disappear because an AI agent enters the workflow.

If anything, understanding those complexities becomes more important as decision-making becomes increasingly autonomous.

The future of agentic payments won't be defined solely by what AI agents are capable of doing. It will be shaped by how effectively we translate those capabilities into secure, compliant, reliable payment experiences and how successfully we bring people along in that transformation.

Michelle Brahmbhatt (Kuar)
Founder & Principal Consultant
Kuar Consultancy