Get the hottest Fintech Hong Kong News once a month in your Inbox
Checking someone’s ID at the door of a nightclub tells you who they are, but it does not tell you how they will behave once they are inside.
Lionel Grosclaude, CEO of Fime, used that analogy to explain a challenge emerging in agentic commerce, where AI agents are beginning to search, shop and make payments on behalf of consumers.
The comparison comes as agentic commerce moves rapidly from concept to live infrastructure.
The Nightclub Test for AI Payments Fime’s FACT Framework makes the case for continuous checks in AI payments, because an agent can still go off track after it is approved. fintech AI commerce payments
Worldpay research also suggests that consumers are becoming more comfortable with the idea, with 73% globally open to using AI agents to browse and purchase on their behalf, rising to 85% in Singapore and 95% in China.
Why Payment Approval No Longer Tells the Whole Story
The size of the opportunity of agentic commerce also explains why the trust problem cannot be treated as a small technical detail.
Lionel linked that projected scale to today’s dispute rates, pointing out that even a 1% dispute rate in a US$3 trillion to US$5 trillion market could turn into tens of billions of dollars in contested transactions.
Agentic commerce starts to look different from ordinary e-commerce when a successful payment no longer proves the consumer got what they meant to buy.
Authentication can confirm who initiated a transaction, and payment networks can move money, but neither was designed to determine whether an AI agent has remained faithful to the user’s original instruction.
Which leaves the industry with a question hanging over their heads.
Once an AI agent starts acting on behalf of a consumer, how can merchants, banks and payment networks know that the final purchase still reflects what the consumer actually intended?
A Payment Can Be Correct and Still Be Wrong
Lionel’s starting point was that the internet, and by extension much of digital commerce, was built around human action.
A person searches, compares and decides before clicking the button to pay.
Agentic commerce changes that sequence because the consumer may still give the original instruction, while the selection and purchase decision happens through software acting in between.
Existing payment safeguards can still confirm who is involved in the transaction. They do not, however, answer whether the agent stayed within the boundaries the user set.
As Lionel put it, the industry will need to trust that the agent is “staying within its mandate.”
Know Your Agent, or KYA, has started to emerge as one way to check an AI agent when it is onboarded.
Fime’s CEO sees value in that approach, although he warned that an onboarding check only captures one point in time.
Lionel Grosclaude
“You can test it at the beginning,” he said, but that does not mean the agent will not “adapt, learn and at the end of the day behave in a different way.”
The risk becomes easier to understand once the agent starts buying real products.
“It could be a simple mistake … you ask for blue shoes, and he buys you a red pair of shoes. This is an issue, but it’s not that serious,” Lionel said, putting the problem in everyday terms.
A merchant may process the order exactly as requested and still leave the consumer with the wrong item.
Health or safety raises the stakes further because one missed allergy can turn a mistaken purchase into something harmful.
The system may still treat the transaction as valid, but that matters little if the agent has already bought something the user should never have received.
When a merchant receives a request from a shopping agent, the framework can compare that request against the consumer’s original prompt, helping determine whether the purchase remains consistent with what the consumer intended.
Lionel described the process as checking whether “what the agent is asking me is totally consistent with the intent of the human.”
The limits of onboarding also explain why Lionel does not see Know Your Agent as the full answer.
KYA can help check an agent when it first enters the system, but it does not prove how the agent will behave later when real transactions begin.
Agent protocols can help systems communicate, while payment networks can move the money.
FACT focuses on the missing link between those functions: whether the agent’s request still matches the consumer’s original intent at the moment of purchase.
In his explanation, the merchant agent asks the FACT agent to check the intent. FACT then accesses the original human prompt and compares it against what the shopping agent is requesting.
If the request is consistent, the merchant can proceed. If not, the merchant can push back before placing the product in the basket.
The transaction-level check matters because it happens when the agent is trying to buy something, not only when the agent is first approved.
Lionel’s earlier nightclub analogy helps explain the difference. Checking someone’s ID may be necessary, but the harder question is whether they continue behaving properly once they are inside.
FACT, in his view, is closer to making sure the agent continues “playing by the books” after the initial check has already happened.
Fime Wants FACT to Sit Outside the Platforms
As agentic commerce spreads across separate ecosystems, Lionel sees a risk in leaving intent checks to each platform.
Protocols may help agents communicate and transact, but as he put it, “they’re not checking the intent.”
“The platform has to work with any protocol,” he said. “It has also to work with any payments network.”
A neutral layer becomes more important as agentic commerce fragments across AI platforms, payment networks, merchants and banks.
Without shared trust signals, every ecosystem may end up checking agent behaviour in its own way, making disputes harder to resolve and standards harder for merchants to navigate.
Merchants should not need a different intent-checking model for every agent, wallet, payment network or platform they support.
The more fragmented the market becomes, the more important it becomes to have a way of verifying intent that can sit across those systems.
Building Evidence Before Disputes Happen
Lionel’s argument returns to disputes because agent-led purchases could create a commercial problem long before the market reaches full scale.
If agentic commerce reaches the size forecast by McKinsey and Bain, even a small dispute rate could create a large exposure.
The problem lies harder to resolve when each party sees a different part of the transaction.
A consumer could argue that the agent misunderstood the instruction, while the merchant could point to the request it received and fulfilled.
Banks and payment networks may only recognise the transaction as valid, while regulators may ask how the agent made the decision and whether the right safeguards were in place.
FACT helps create the evidence needed before a dispute happens.
“If something goes wrong, and if the merchant has used FACT, he is going to be able to have a report,” Lionel said.
What If AI Makes a Bad Purchase For You? AI can already shop and pay on your behalf. But what happens when it buys something you never wanted, or worse, something that puts you at risk? fintech AI payments
The report could show whether the transaction requested by the shopping agent was consistent with the user’s intent.
Such a record matters because liability in agentic commerce will not always be obvious.
A human checkout creates a relatively direct link between decision and payment, while agentic commerce inserts another layer into that process.
Before the payment reaches the merchant or bank, the agent may have already made the purchase decision on behalf of the consumer.
FACT attempts to make that boundary clearer by preserving evidence around what the agent was allowed to do and whether it followed that instruction.
Merchants could find that protection increasingly important. Lionel said a merchant that cannot prove it respected the user’s intent may find itself responsible when the consumer challenges the transaction.
Regulators Will Need More Than Promises
The need for evidence also matters because regulators are bringing AI governance further into financial services.
Take, for example, the EU AI Act, which has raised expectations for how firms govern higher-risk AI systems, while Singapore’s FEAT principles have long pushed financial institutions to consider fairness, ethics, accountability and transparency in their use of AI and data analytics.
Lionel said FACT can tune its checks according to the rules of the market where the transaction occurs, rather than apply one fixed standard everywhere.
Agentic commerce may require firms to prove on an ongoing basis that they are checking autonomous transactions against the rules in each jurisdiction, instead of treating compliance as something completed at launch.
Regulators are unlikely to accept vague claims that an AI agent was trusted.
They will need to understand what the consumer authorised, what the agent requested, how the merchant responded, and whether the transaction stayed within the relevant rules.
According to Lionel, FACT provides a check that can adjust to those regulatory expectations while still operating across different markets and payment environments.
Trust Could Decide If We Should Adopt Agentic Commerce
Agentic commerce may already be technically possible, although broad adoption is still far from guaranteed.
Fime’s CEO said relatively few people have actually used AI to buy things so far, even as the use case moves closer to everyday commerce. In his view, trust will decide whether consumers and businesses move beyond early experimentation of agentic commerce.
“The trigger for this market will be trust,” Lionel said.
The line brings the interview back to its central point.
AI agents may soon be able to handle much of the shopping and payment journey on behalf of consumers. The real test is whether the industry can prove those actions still reflect the limits set by the human behind them.
Wider adoption will depend on whether agents remain within those limits, and whether merchants, banks and regulators can prove what happened when something goes wrong.
Fime’s FACT Framework is one attempt to answer that problem before agentic commerce reaches mainstream scale.
So, speaking of agentic commerce, would you trust an AI agent to pay on your behalf, or would you still want proof that it bought exactly what you asked for?
Watch the full conversation with Lionel Grosclaude, CEO of Fime, to hear how the industry may begin answering that question.
