After several months of research, 16 banks representing 5 European countries (Belgium, France, Germany, the Netherlands and Spain) approved the official launch of a new European payment scheme called the European Payments Initiative (EPI) on July 2nd 2020. In this series, we explore the purpose of the initiative and how it plans to achieve its ambitious goals.
1. Introduction
Welcome to the next chapter in financial services. A chapter not about better apps or faster cards, but about software that shops, negotiates, and pays, all without a human lifting a finger. Welcome to the age of agentic commerce.
Now, let’s not mince words: this is not your standard “robots will steal our jobs” fable or Silicon Valley’s latest fever dream. This is serious and imminent
It is a world in which AI agents not only advise on purchases but are given the keys to the financial kingdom to execute them. And not just once. They do it all the time, in the background, tirelessly and autonomously.
Imagine a Siri with a bank account, or Alexa negotiating with another bot to buy cloud computing credits at midnight while you sleep. This is not some half-baked science fiction concept. It is Visa, Mastercard, PayPal, and Stripe laying the foundations today.
So let’s unpack what this really means, what is at stake, and what we must do to prepare, not only to survive but to thrive in a future where software becomes a bona fide market participant.
2. Agentic commerce: the elevator pitch with an AI concierge.
Let’s begin with the basics. Agentic commerce refers to economic activity initiated and executed by autonomous AI agents. Not decision support systems. Not fancy dashboards. Proper agents, with their own logic, permissions, goals, and, critically, access to money.
These agents come in two main flavors:
Via-agent payments: the AI acts on your behalf, a glorified concierge. You are still in charge, but the agent does the legwork: “Find me a flight to Berlin under £250 and book it.” Sorted.
Inter-agent payments: machine-to-machine. Your electric vehicle pays the charging station. Your fridge subscribes to a cheese-of-the-month club. Your drone negotiates airspace priority fees with a municipal traffic AI. These are payments between autonomous economic agents, not simply automated billing.
This is not automation. This is autonomy. Software becoming a transacting entity with identity, credentials, permissions, and a wallet.
3. From plastic cards to silicon minds.
We have seen this evolutionary pattern before. First came cash. Then plastic. Then the mobile wallet. Each iteration did not just shift user behavior, it transformed the underlying economics of commerce. Agentic commerce is that transformation taken to its logical extreme.
The leap here is qualitative, not merely incremental. An AI agent that can act, pay, and respond in real-time introduces a radically new type of actor into the marketplace. It is not just that commerce is faster or more convenient. It is that humans are no longer always in the loop.
And the big guns are already positioning. Visa’s “Intelligent Commerce,” Mastercard’s “Agent Pay,” Stripe’s “Agent Toolkit,” Google’s AP2 protocol, OpenAI’s ACP protocol and Coinbase’s AI wallet infrastructure are all signals of a paradigm shift.
A good analogy: think of the shift from dial-up to broadband. At first, the web was a thing you visited. Then it became ever-present. Now imagine commerce is the same. The AI agent is not a tool you use. It is a partner, perpetually transacting.
4. Trust me, I’m an AI (with a wallet)
Now we arrive at the age-old question: who do you trust?
In “Identity is the New Money”, Dave Birch argued that in a digital-first economy, identity is more fundamental than money. And here, that thesis becomes manifest. You cannot allow autonomous agents to pay unless you know who they are, what authority they have, and who they ultimately represent.
KYA - Know Your Agent.
Enter the concept of KYA, Know Your Agent.
It is like KYC’s slightly weirder cousin, but it is vital. Imagine your AI assistant gets hacked and starts sending micropayments to a rogue actor’s machine farm in Moldova. Who is responsible? Who stops it? Who gets the refund?
To answer these questions, every AI agent will need to carry:
A verifiable digital identity (anchored to a legal entity or individual).
A smart wallet (with programmable constraints).
A reputation record (built on past behaviour and network signals).
Different players are taking different routes. Google is leaning on verifiable credential. OpenAI is building a more controlled ecosystem closely integrated with PSPs. Mastercard and Visa are enhancing their tokenization services to enable agent-based payments.
Identity alone, however, is not enough.
Measures to mitigate AI nondeterminism.
Even if you know your agent, it does not mean your agent is right. AI systems are inherently probabilistic. They interpret intent, weigh tradeoffs, and make judgments that may be technically rational yet contextually wrong.This introduces the need for additional safeguards, such as:
Explicit delegation scopes tied to defined intents.
Spending limits linked to categories and risk appetite.
Real time monitoring and anomaly detection tailored to agent behavior.
Revocation mechanisms that can immediately suspend authority.
Audit trails that explain why a decision was made.
Think of the agent like a corporate intern with a prepaid card. They have rules. They are monitored. They operate within limits. And they are accountable, sort of.
5. Smart wallets: where the magic happens.
The humble wallet is no longer just a repository for tokens. It is a policy engine. A governance container. A programmable boundary between autonomy and compliance.
Smart wallets for AI agents will carry not only digital money (tokens, fiat, CBDCs, etc.) but also delegation logic:
Spend limits.
Merchant restrictions.
Risk flags.
Behavioral rules.
Regulatory triggers.
A merchant, in turn, may only accept payments from agents with valid credentials, certified wallets, and a trust rating above a threshold. Welcome to algorithmic commerce with built-in governance.
And yes, that opens the door to programmable identity-linked reputation systems. Your bot might be banned from a network for late payments or suspicious behaviour, just as you can be blacklisted from Airbnb or Uber.
The future is one where bots build (and burn) credit reputations like humans do.
6. The institutional headache.
Now, if you are a policymaker or a central banker, you should put down your policy memo and picking up a stiff drink. Because the implications of agentic commerce are seismic.
6.1 Legacy infrastructure ≠ AI-ready
The current payment rails – even the fancy new ones like FedNow, SEPA Instant, or UPI – were built for humans. Humans who sleep, who misplace their card, who take weekends off.
AI agents? They transact at millisecond speeds. They do not pause. They do not make typos. They scale. If you think fraud detection is hard now, imagine monitoring billions of machine-initiated micropayments per hour.
6.2 Responsibility vacuum
Who is liable for a faulty agent transaction? Is it the developer? The wallet issuer? The API provider? We are entering a legal fog where the old rules – like consumer protections, chargebacks, and dispute arbitration – do not map.
We will need new legal frameworks. Perhaps a concept of “digital power of attorney.” Or “algorithmic fiduciary duty.” The key is clarity without stifling innovation.
6.3 AML and sanctions compliance go out the window
Bots do not carry passports. They do not show up at branch offices. Yet they will be sending money cross-border. How do you ensure AML compliance, sanctions enforcement, and tax reporting?
Traditional AML is predicated on human profiling. What does it mean to profile a bot? Behavioural biometrics? Logic audits? Source code transparency?
You tell me. But someone is going to need to invent the RegTech stack for the agentic era.
7. The strategic imperative: prepare or perish.
The agentic commerce revolution is not a fringe phenomenon. It is a tectonic shift akin to the rise of the web or the smartphone. If central banks and domestic payment schemes do not adapt, they will be bypassed, not maliciously, but efficiently.
The agentic world demands:
24/7 always-on networks.
Real-time settlement.
Support for micro- and nano-transactions.
Programmable identity layers.
Bot-aware fraud detection.
Some may argue that Distributed Ledger Technology (DLT) offers an answer – and in some use cases, it might. But let’s not confuse the means with the ends. The goal isn’t blockchain. The goal is resilient, auditable, AI-native payments infrastructure.
8. So what should we do?
8.1 Short-term (next 1–2 years)
Agent payment sandboxes: regulatory sandboxes specifically for agentic payments. Think FCA meets AI zoo.
KYA protocols: standardize digital identities and credentials for AI agents.
Virtual payment credentials: banks and wallets should offer “agent cards” with limited privileges and traceable use.
Message standards update: update ISO 20022 and other formats to include “agent flag” metadata.
8.2 medium-term (3–5 years)
Programmable CBDCs: central banks should pilot programmable digital currencies that can be used by AI agents under smart contract rules.
New legal definitions: introduce legal constructs for agent delegation, algorithmic liability, and dispute resolution.
Agent licensing: just as payment providers need licenses, so too may certain classes of autonomous agents.
8.3 long-term (5+ years)
Redesign payment infrastructure: build a bot-native core. Fast, scalable, transparent, and fraud-resilient.
Cross-border bot payments: create agent-compatible standards for cross-border interoperability – an “IBAN for AI,” if you will.
Global agent registries: an international body (perhaps BIS or IMF) should coordinate registries of verified AI economic agents.
9. Conclusion: we have seen the future, and it has a bot ID.
With agentic commerce, money does not just understand you, it acts on your behalf.
Money is becoming ambient, anticipatory, and autonomous. Not because the technology is cool (though it is), but because the economics of agency demand it.
The number of decisions a consumer makes each day is growing. Choice overload. Friction fatigue. Delegation is inevitable. The winners will be the platforms –and the economies – that make delegation safe, efficient, and trusted.
We are witnessing the birth of a new class of economic actors: bots with budgets. If our payment systems cannot embrace them, someone else’s will. Probably in a jurisdiction with fewer lawyers and more vision.
The agentic era has begun.
The only question is: are we ready to give AI a wallet, and trust it not to buy 10,000 NFTs of pixelated squirrels?
Let’s hope so. Because the alternative is sitting on the sidelines while your toaster negotiates a power contract behind your back.
Raphaël Guilley, SVP Strategic Portfolio and Growth
Raphaël has over 20 years of experience in the consulting industry, with extensive involvement in managing large-scale international projects across payments, smart mobility and digital identity. His areas of expertise include product management, agile development and product launches.
At Fime, Raphaël leads the global Consulting team under the Consult Hyperion brand, following Fime’s acquisition of the company. He supports a wide range of stakeholders, including payment networks, financial institutions and transport operators, to solve complex challenges, explore new opportunities and expand into new markets.
Prior to joining Fime, Raphaël was VP of Risk & Compliance Solutions at IPC Systems Inc. He also worked in similar roles for Etrali Trading Solutions and Orange Business Services.
