For the past two years, when people talked about artificial intelligence, the imagery always ran to the same images: chatbots churning out texts in perfect human style, image generators challenging photographers and designers, language models responding as conversational encyclopedias. All true, but that was not where the most radical transformation was maturing. While public attention remained entranced by the more spectacular aspects, something quieter, and perhaps more disruptive, was taking shape in the operating rooms and investment committees: the arrival of the AI agents, systems capable not only of grinding out data, but of taking action, interacting with other software, learning from their mistakes and - within limits - making autonomous decisions.

The difference with the first generation of AI is substantial. Machine learning algorithms of the past decade have revolutionized the ability to analyze time series, detect anomalies, optimize portfolios, or unearth patterns in markets. But they remained supporting tools, serving a human analyst who guided their moves. Intelligent agents, on the other hand, shift the center of gravity: they can constantly monitor information flows, adapt strategies, execute trades, and even "talk" to other systems. Not just software, but a kind of digital colleague participating in decision making.

The first serious experiments took place in the territories where speed is everything: the algorithmic trading and the halls of hedge funds. Here AI agents are trained to react in milliseconds to price changes, macroeconomic releases or even alternative data such as nighttime lighting picked up by satellites. Rather than mere executors, they become tireless opportunity hunters: testing micro-strategies, learning whether they work, discarding or refining them. In an ecosystem where time is measured in microseconds, the possibility of a system learning and adapting its behavior without waiting for the hand of the analyst represents a paradigm shift.

Yet it is in the wealth management that these agents are showing the most transformative face. Private banking has been living under strain for years: increasingly thin margins, stringent regulations, increasing demands for personalization. A combination difficult to manage with human teams alone. Digital agents open the door to a different model. Imagine a family office with dozens of complex mandates: portfolios that must meet ESG criteria, cross-border tax constraints, currency exposures to cover, plus wealth succession needs. A programmed and supervised AI agent can become the invisible director: it collects data in real time, flags unexpected correlations, and proposes immediate rebalances, minimizing the latency between event and decision.

The role of the banker does not disappear, but it changes its skin. From an operator who performs analysis and proposes choices, he becomes an interpreter and supervisor, the one who translates the machine's proposals into the client's language and vice versa. A liberation from repetitive tasks, leaving more room for the heart of the profession: the fiduciary relationship.

In some pilot projects already launched in Europe and Asia, clients can talk directly with a digital agent. Not the usual robo-advisor offering pre-packaged portfolios, but an interlocutor who answers sophisticated questions in natural language: "What happens to my bond portfolio if the ECB raises rates by 50 basis points?" or "What is the tax treatment if I move part of the trust to an Asian jurisdiction?" The agent combines market data, regulations, and personal profile, providing tailored scenarios. A level of customization that until a few years ago required senior consultants and hours of work.

Here, however, the first delicate knots emerge. The most immediate one concerns the trust. Are those who own family assets accumulated over generations willing to rely on a digital entity? The history of finance suggests that technologies only gain traction when flanked by a robust system of accountability and oversight. And indeed, regulators are moving: the SEC in the United States, ESMA in Europe, and FINMA in Switzerland have begun to discuss whether and how these tools should be regulated. Are they mere extensions of advisory services, or new entities that require specific rules?

The second node touches the very identity of the profession. Some commentators foresee radical disintermediation, with agents capable of replacing a substantial proportion of bankers. Others, more realistic, see the emergence of a hybrid model: the machine as the analytical and operational engine, the human being as the guardian of the relationship and guarantor of trust. After all, wealth management has never been just about numbers: inside are political sensitivities, personal values, family tensions. Dimensions where empathy, at least for now, remains beyond the reach of any algorithm.

Then there is the problem of thealgorithmic opacity. If an agent recommends liquidating an asset or taking on a complex hedge, the client and-inevitably-the regulator will want to understand why. This is the theme of "explainability": explaining how and why a system made a decision. Already discussed in AI applied to health care or justice, it becomes crucial in finance, where fiduciary responsibility does not allow for gray areas. Some developers are introducing "explanatory dialogue" features, which allow the user to hold the machine's choices accountable. This is a step forward, but not yet the ultimate solution.

At the geopolitical level, three distinct models emerge. In the United States, the pragmatic, fast-track approach prevails: hedge funds and fintechs are already in the field, driven by venture capital and looser regulation. In Europe, caution dominates: large universal and private banks test prototypes in controlled environments, under the watchful eye of the authorities. China is proceeding on a different track: integrating intelligent agents into financial conglomerates and domestic big tech, crossing efficiency goals with logics of state control of data. Three paths that reflect three ways of understanding the relationship between innovation, risk and power.

It is hard to say how far we will go. It is plausible that within a few years we will see agents capable of orchestrating global multi-asset portfolios, negotiating bespoke derivatives, optimizing philanthropic strategies or even managing sovereign wealth funds' assets. But the real game will be played elsewhere: in the ability to build a balance between autonomy and oversight, between efficiency and transparency. AI agents may reshape the physiognomy of wealth management, but one question remains that no technology alone will be able to resolve: in wealth governance, who will really remain at the helm, man or machine?