At the start of June, Money20/20 returned to Amsterdam. As it does every year, the global payment industry’s flagship event¹ brought together banks, fintechs, payment networks, regulators, investors, and tech giants for three days to discuss the sector’s major transformations. But this year, the atmosphere has changed.

Whereas previous editions were dominated by the development of new payment channels, the expansion of open finance, or the emergence of new tokens, 2026 has put artificial intelligence front and center, leaving an entire ecosystem in a state of shock.

With AI and its army of agents, a world is emerging where tomorrow’s economic decisions promise to be increasingly delegated to machines… without us truly knowing what agent-based systems are or will be!

If we are to believe AI advocates, paying could soon disappear behind intentions expressed in natural language.

Buying, comparing, deciding, or paying—all actions humans can forget about, because tomorrow, an AI agent could handle them for us. In this vision, payment would become almost background noise, a technical gesture relegated to the background of a truly automated experience.

This would be the pinnacle of payments. And yet, it is precisely because payment risks becoming invisible to the user that it must become much more visible to the financial industry itself.

Visible in its origin. Visible in its authorization. Visible in its traceability. Visible in its accountability. Visible, finally, in the data that allows an agent to understand what the user really wants to do.

For behind the promise of an AI-driven world lies a far more complex reality. Agent-based AI does not merely add a conversational interface on top of existing payments. It shifts the very center of gravity of the industry, which, when executing a transaction, will have to determine who authorized it, on whose behalf, based on what data, with what limits, under what accountability, and within what framework of trust.

The payment industry is thus entering a paradoxical phase. The more fluidity it promises the end customer, the more it will need to strengthen control mechanisms behind the scenes. The simpler the experience becomes, the more sophisticated the infrastructure must be. The more autonomous payments become, the more central the question of liability will become.

In recent years, innovation in payments has largely focused on reducing friction within processes. Fewer clicks, fewer data entry fields, fewer abandoned carts, and fewer interruptions in an online customer’s journey, for example. Overall, payment has gradually blended into a seamless, carefully designed shopping experience.

But AI agents are turning everything upside down. In a traditional model, embedded across multiple environments, the user browses, compares, chooses, confirms, and then pays. In an agent-based model, they formulate an intention within a new framework. They no longer simply say, “I want to pay for this product.” They can instruct their preferred AI: “Find me the best option,” “Stay within my budget,” “Prioritize this criteria,” “Avoid this merchant,” “Renew this service if the terms are favorable.”

This shift is significant. Commerce would therefore no longer be centered on the interface, but on intent. Yet intent is rarely simple. It is contextual, evolving, and sometimes contradictory. It combines explicit preferences, implicit constraints, financial trade-offs, consumer habits, a relationship to risk, a need for security, and sometimes even an emotional dimension.

In this dizzying new paradigm, paying amounts to executing a decision made—or prepared—by an autonomous system. So, did the agent truly understand what the user wanted? Did it respect the user’s constraints? Did it optimize in the user’s actual interest or only according to overly narrow criteria? Did it act within the proper scope of delegation? Did it rely on another agent, itself controlled by a third party? Agent-based payment raises the question of mandate.

It would, however, be reductive to limit the topic of agent-based systems in the world of payments to the checkout moment alone. In the corporate world, payment is by nature rarely an isolated act. It is part of a longer chain: order, invoice, validation, reconciliation, settlement, reporting, cash management, risk control, accounting.

The AI agent thus fits into a complete economic process. And that is likely where part of the value lies. Payment should no longer be viewed in isolation, but rather in terms of its ability to be orchestrated with the business processes surrounding it.

We must avoid viewing agent-based commerce through the narrow lens of the transaction itself. The real issue ultimately boils down to this question: can an agent understand, initiate, and document a complete economic transaction?

The financial industry is well acquainted with KYC, the process by which an institution identifies its client, verifies their identity, identifies their ultimate beneficiaries, and regulates the relationship. But what happens when a transaction is no longer directly initiated by a human, but by an agent acting on their behalf? This is where an idea is emerging that is set to gain importance: Know Your Agent.

In the future, it will no longer be enough to know your customer. You will also need to know your agent—or the army of agents behind them. You will therefore need to identify this chain of agents. This means authenticating them, understanding who developed them, which organization they are affiliated with, what permissions have been granted to them, what actions they can perform, on what amounts, under what circumstances, and with what revocation options.

This issue goes to the heart of trust in payments. An agent capable of purchasing, transferring, subscribing, renewing, or arbitrating becomes an operational player in the value chain.

The ability to distinguish between a legitimate user, an authorized agent, and its malicious twin could thus become one of the major challenges of the digital economy. In a world where agents interact with one another, sometimes monitor each other, and perform tasks at high speed, identity can no longer be viewed solely from a human perspective.

KYA could therefore become in the agent-driven era what KYC was in the banking era: a trust infrastructure, but also a formidable barrier to entry.

Trust, however, cannot rest solely on a promise, a brand, or a reassuring interface. In agent-based payments, there will no longer be blind trust. It must be verifiable at the very moment money moves – or even before.

This requires a much more explicit delegation architecture. Which agent is acting? On behalf of which user? Within what scope? With what mandate? For how long? Up to what amount? With what level of human confirmation? And with what proof that the initial intent was indeed respected?

The vocabulary heard throughout the exhibition hall is telling: permissioning, spending limits, step-up confirmation, signed intent, policy engine, audit trail, delegated authority. In other words, trust will have to be coded, signed, limited, monitored, and audited to be contestable.

This is one of the most important issues for the financial industry. We cannot allow an agent to make financial transactions on a whim, driven by delusions, for example. A simple instruction in a prompt is not enough. For agentic systems to scale up in the world of transactions between economic agents, a control layer will be needed that can say no, even when the agent seeks to achieve its goal by any means necessary. Invisible payments will therefore need to produce visible evidence.

Payments belong to a deeply deterministic industry. A transaction must be executed or rejected. An authentication must be validated or rejected. A settlement must be calculated. Liability must be assignable. An audit must be able to reconstruct the sequence of events.

AI, on the other hand, introduces a more probabilistic logic. It infers, recommends, interprets, and anticipates as well. It can produce powerful results, but rarely with the formal rigidity expected of a banking or payment infrastructure.

This is undoubtedly one of the major tensions of the coming years. We will need to make AI systems—which are by nature imperfectly deterministic—coexist with financial processes built on the requirements of certainty and correctness.

This tension does not, however, spell the end of AI in payments. Rather, it sets the conditions for its acceptability. AI agents will only be able to be deployed on a massive scale in financial services if they are governed by robust safeguards: human validation, action thresholds, explicit permissions, the ability to challenge decisions, emergency suspension mechanisms, and reversibility.

The “human in the loop” is therefore not disappearing. It is changing its role while asserting itself more strongly. Humans are no longer necessarily the ones who execute every action. They become the ones who set the framework, supervise, limit, validate, introduce friction and constraints, and take back control when autonomy becomes too risky. In fact, the automation of financial services driven by agent-based systems does not eliminate governance. It makes it even more indispensable.

An AI agent never acts in a vacuum. To be useful, it must understand the context. In payments and financial services, this context is remarkably rich: income, spending habits, market preferences, budget constraints, transaction history, net worth, risk aversion, product, geography, timing of expenditures, banking relationships, savings capacity…

The value of a financial intermediary will therefore depend less on its algorithmic power alone than on the quality of the data it can access. And this is where a new industrial battle is taking shape.

Banks have a relationship of trust and privileged access to part of the financial landscape. At their level, fintech companies capture usage patterns and often offer a more refined user experience. Major payment networks control flows, standards, and massive, in-depth transactional data. Merchants, for their part, have product catalogs, prices, inventory, and purchasing preferences. ERP leaders organize the financial operations of businesses. Together, all these players hold this context, which still eludes the LLM giants who would love to consolidate it. But the reality is that no single player naturally holds the entire context.

Without consolidated, structured, and interoperable data, agents risk producing weak recommendations, poorly calibrated decisions, or inadequate actions. The promise of agent-based systems thus collides with a very concrete reality: financial data that is scattered, incomplete, sometimes inconsistent, and never perfectly neutral.

At Money20/20, industry players are saying it. It’s not enough to have data. It must be readable by agents.

This is a more profound change than it appears. A bank may have a vast amount of information on its customers, products, rates, risks, and transaction histories. A merchant may have a rich catalog, inventory, prices, reviews, and terms of sale. But if this information isn’t structured in a format that agents can use, it risks becoming invisible in the new purchasing journey.

The battle for data is thus shifting from its ownership to its ability to be consumed by agents. In agent-based commerce, an offer that is not understood by the machine is an offer that risks not being presented.

For banks, a financial product that cannot be compared, contextualized, and interpreted by agent-based systems may be bypassed in favor of a more readable one. In this ecosystem, agents will seek out the most readable, most integrable, and most actionable products. Another example: a payment option that cannot be enabled by agent-based systems is an option that risks disappearing from the checkout process operated by machines.

For merchants, for example, this means transforming product catalogs. A marketing feed with just a few attributes will no longer necessarily suffice. Agents will need more context: features, constraints, availability, terms, variants, compatibility, return policies, fees, warranties, delivery times, and trust signals. The transition from a catalog designed for human management to one designed for machine interpretation represents a major infrastructure project.

In the agent-based economy, not being machine-readable could become a new form of commercial exclusion.

This new centrality of data poses another challenge. Financial data is often presented as objective because it is quantified. Transactions, income, balances, due dates, expenses, repayments: everything seems measurable, and therefore neutral.

This neutrality is an illusion. Financial data reflects social, professional, geographic, wealth, and behavioral trajectories. It can carry biases, whether structural or behavioral. A model trained on this data can therefore reproduce, amplify, or normalize asymmetries already present in the current system.

This financial data is sensitive because it can become prescriptive. An agent who recommends a product, adjusts a path, prioritizes an option, avoids a risk, or triggers an action does not merely read a situation. They help transform it. The quality, governance, and explainability of the data will therefore become central dimensions of agent-based payment.

When it comes to instant payments, economic actors experience parallel realities. When the user sees a notification, the merchant sees a confirmation. The app displays a completed transaction. But the payment chain still consists of distinct steps: authentication, initiation, authorization, notification, clearing, settlement, and actual availability of funds. Today, the experience can be instantaneous without the entire infrastructure being fully so.

In a world of AI agents capable of executing actions at high speed, this speed of payments will automatically become a more pressing issue. But speed alone will not be enough. We will still need to know what is truly instantaneous: the order? The notification? The authorization? The clearing? The final settlement? The availability of funds to the merchant?

Tomorrow payment could thus evolve into a continuous flow, a sort of streaming that breaks with certain batch-processing paradigms. But here again, the challenge will not be merely technical. It will also be legal, accounting, operational, and prudential.

What is real-time for the customer may still be delayed for the infrastructure.

In every cycle of financial innovation, the established players are quickly portrayed as the next victims. Stablecoins were supposed to bypass established networks. Open banking was supposed to weaken banks. Fintechs were supposed to disintermediate an entire ecosystem. AI agents, in turn, might lead us to believe that existing infrastructures are about to disappear.

These clichés no longer hold water. Payments constitute one of the most complex industries in digital capitalism. They encompass regulation, compliance, technical standards, fraud prevention, customer experience, risk management, interoperability, merchant relationships, clearing infrastructure, and resilience requirements.

In such an environment, the execution capabilities of established players remain a powerful advantage. Visa, Mastercard, major banking networks, market infrastructures, and certain regulated account-to-account service providers are not merely conduits. They are architectures of trust, distribution, and accountability.

Agent-based systems do not necessarily spell their doom. They could even strengthen their role if these players manage to become the trusted orchestrating of a multi-rail, multi-agent, and multi-data world in the era of AI’s staggering power.

The real issue, then, is not whether new protocols will replace old networks, but rather who will be able to orchestrate their coexistence.

The great promise of modern payment has long been its invisibility. No longer having to think about it. No longer having to pull out your card. No longer having to enter your information. No longer having to wait. No longer experiencing friction. Tomorrow, perhaps, no longer having to decide on every micro-action.

But this apparent disappearance is a fantasy. Payment never truly fades away. It shifts. It migrates from the click to the intention. From the user to the agent. From the invisible journey to the visible layer of trust.

This is where the next phase of the payments industry will play out. Not in the naive and simplistic promise of a world where agents would freely pay on our behalf, but in the patient orchestration of a sustainably fragmented environment where they can be identified, limited, supervised, challenged, and understood. Without this, invisible agent-based payment will likely never scale.