The Agentic Age Arrives in European Finance
At Money20/20 Europe, the conversation shifted from what AI might do to what it is already doing
For three days in early June, the Money20/20 Europe gathering in Amsterdam played host to more than 7,600 delegates from over 100 countries, and the dominant subject was not a product or a payment rail but a posture. The industry has stopped asking whether artificial intelligence can assist financial work and started asking whether it can do that work on its own. Of the four themes the organisers chose to frame the event, the one that coloured almost every panel was "AI and the Agentic Age." Bryony Naylor, who was promoted to vice-president of Europe at Money20/20 last autumn, captured the mood in her opening remarks.
"AI is moving from experimentation to execution, stablecoins are entering the mainstream, regulation is accelerating, and every leader here is asking 'who will build what comes next.'"
The distinction between generative and agentic AI matters here, because it changes what is at stake. Generative systems produce text, images and code for a human to review and use. Agentic systems perceive a situation, plan a sequence of steps, and then act: initiating a payment, adjusting a credit limit, rebalancing a treasury position, or chasing down a suspected fraud without waiting to be told. In finance, where actions move money and create liability the instant they execute, that autonomy is both the promise and the problem.
A competitive advantage measured in revenue per employee
The clearest articulation of the opportunity came from a panel on how to build the next ten-billion-dollar fintech, where Antony Jenkins, the former Barclays chief executive who now runs the core-banking firm 10x Banking, argued that scale will be won or lost on how well a company puts agents to work.
"The defining source of competitive advantage in the next five years is going to be the ability of the organisation to learn how to deploy these agents effectively."
Jenkins went further, suggesting that agentic systems could see a company's revenue per employee double or triple. That figure is the kind of claim that sounds like conference theatre until one considers where it would come from. Banks and fintechs run armies of people through multi-step processes that are mostly rules and judgment: reconciling ledgers, triaging fraud alerts, gathering documents for a credit decision, answering the same compliance questions in slightly different forms. An agent that can complete those chains end to end, rather than merely drafting one step for a human, removes the labour from the loop rather than speeding it up. In payments the same logic points toward self-optimising routing, autonomous reconciliation and real-time decisioning that adjusts as conditions change.
Evidence that vendors are building for this future arrived during the event itself. Experian used its Amsterdam slot to launch an Agent Operating System, a layer designed to orchestrate AI agents across fraud detection, identity verification and credit-risk workflows, with ServiceNow signed up as its first partner. The architecture is telling: it ships with model-risk monitoring and human-in-the-loop controls for high-stakes lending built in, an acknowledgment that autonomy in finance has to come wrapped in governance or it will not be bought.
Governance moves upstream
If the opportunity dominated the startup stage, the harder questions surfaced in a session devoted to the regulatory realities of agentic commerce. Nicole Sandler, chief ecosystem officer at the clearing system Ubyx, made the case that automated transactions break the assumptions on which compliance has long rested. In conventional retail, the controls sit at the checkout. When an agent is researching, negotiating and buying on a customer's behalf, the decisive moments happen much earlier in the chain.
"The problem is very much that governance has moved upstream."
Sitting alongside her, Colin Payne, head of innovation at Britain's Financial Conduct Authority, agreed that the timeline has compressed. The regulator has been preparing for autonomous agents in the abstract, and now finds them in the room.
"Where I was predicting it to be coming down the track, now it's arrived."
Colin Payne, Head of Innovation, Financial Conduct Authority
Payne described the FCA as a tech-positive regulator focused on consumer outcomes, and pointed to the agent-based use cases it is already running inside its AI Lab. Sandler's warning was that none of this scales without two foundations the industry has barely begun to build. The first is a clear allocation of machine liability, so that when an agent errs there is a defined answer to who pays. The second is what she called a "know your agent" framework, an analogue to know-your-customer rules that lets one system verify the identity and authority of another before transacting with it. Without those, she argued, there will be no real scale for agentic commerce or agentic payments. Payne tied the point to standardisation, noting that the FCA is testing use cases through the Global Financial Innovation Network alongside seventeen international regulators, on the view that technology without a standardised process tends to fail.
Europe's particular bet
That emphasis on standards and oversight is where Europe's position becomes interesting rather than merely cautious. The region's AI Act already classifies systems that assess creditworthiness as high-risk, which means documented risk management, bias-tested data, audit logs and genuine human override on consequential decisions. The headline compliance deadline of August 2026 looms over every credit model in production, though a proposed Digital Omnibus package could push parts of it toward late 2027. Prudent institutions are planning as if the earlier date holds.
The strategic question is whether this regulatory weight is a handicap or a moat. A continent that defines machine liability, agent identity and human oversight before the technology saturates the market could export those standards the way it exported data-protection norms, and could offer customers a version of agentic finance that carries trust as a feature. The same rules could also slow European firms while less constrained competitors in the United States and China move faster. The agentic interface raises the stakes by threatening the bank's relationship with its own customer. If people come to delegate their financial lives to an agent, whoever owns that agent owns the interface, and traditional institutions risk being reduced to plumbing behind someone else's assistant.
For all the urgency in Amsterdam, the deployment data counsels patience. Surveys this year suggest that only a small share of enterprises have moved agentic AI into scaled production, with the vast majority of pilots stalling on data readiness, governance friction and uneven model reliability. Financial services lead the field, yet leading the field still means most institutions are experimenting rather than running. The lesson of the conference was not that the agentic age has been won, but that it has begun, and that the firms most likely to prosper are those that pair speed of deployment with the governance to make autonomy safe. As Payne put it, a process that is not standardised fails. In European finance, the race is now to standardise before the agents outrun the rules.