Agentic Investing
Agentic Investing refers to investing in which a computer program doesn't just make suggestions, but independently plans and carries out steps – from research to placing the order. The human sets the goals and boundaries, while the program handles the individual tasks.
In Agentic Investing, a computer program takes over several steps of the investment process itself. It searches for information, evaluates it, draws conclusions, and can ultimately even trigger a buy or sell order. The difference from an ordinary investment program lies in its autonomy: it doesn’t wait for every single command, but works toward a predefined goal. A human might say, for example: “Find me shares of European companies with stable profits and invest 200 euros every month.” The program breaks this task down into subtasks and carries them out one after another. The word “agentic” comes from the English term for someone who acts, meaning someone who takes action of their own accord.
Why banks are betting on it – and why regulators are getting nervous
A large part of an analyst’s job consists of reading. Annual reports, news, data series, analyst commentary – a human can only get through a fraction of this per day. A program can go through thousands of documents in minutes. Large banks and fund companies expect this to bring speed and lower costs above all. Small investors are also meant to benefit: advice that used to be available only to wealthy clients is suddenly becoming cheap to offer.
At the same time, a new risk emerges. If a human makes a mistake, it affects one portfolio. If an automated system has a flaw in its reasoning and is deployed thousands of times over, it affects the market. Regulators also warn of herd behavior: if many providers use similar models, they could all want to sell the same thing at the same time. It is precisely such uniform movements that amplify market crashes.
On top of that comes the question of liability. Who pays if an agent places a wrong order? So far, the answer is clear: the company that operates it. That’s why most systems are, in practice, set up more cautiously than the marketing might suggest.
From instruction to order: the chain behind it
At the core is usually a language model – a program that has learned from vast amounts of text to understand and generate language. On its own, it could only produce text. It becomes useful through so-called tools: access to price databases, news archives, calculation programs, and the broker’s trading interface. The model decides which tool to use and when.
The process runs in loops. The system formulates an intermediate step, executes it, examines the result, and plans the next step. One example: first retrieve the balance sheet figures of twenty companies, then filter out the ten weakest, then check current news on the remaining ones, then write a recommendation. If a step leads nowhere, the system tries a different approach.
Almost everywhere there are fixed limits that the program is not allowed to exceed. Typical examples include maximum amounts per order, bans on certain types of securities, and an approval button for the human before actual execution. Experts call this “human in the loop,” meaning the human serves as a checkpoint. A pure asset manager without any human oversight is currently barely feasible under German regulatory law.
Where the term shows up
It appears most often in reports about banks and brokers. Large institutions such as JPMorgan or Goldman Sachs regularly report on internal assistance systems for their analysts. Neobrokers and fintech firms, in turn, advertise features that monitor portfolios and make suggestions when deviations occur. Often, though, behind the marketing term is nothing more than a chatbot with price data.
A second place the term crops up is in stock market news itself. Providers of trading software, data vendors, and cloud companies cite Agentic Investing as a growth area to make their own shares more attractive. Anyone reading such news should check whether a product actually exists or has merely been announced.
For private investors, a sober assessment is worthwhile. A robo-advisor that rigidly rebalances according to fixed rules is not yet an agent – it follows a rigid plan. It only becomes agentic once the system itself decides which steps are necessary. And even then, what applies to any investment still holds true: if you don’t understand why something is being bought, you shouldn’t buy it.