Agentic AI, a new class of artificial intelligence capable of autonomous decision-making, is beginning to reshape the asset management industry. Dutch investment firm Robeco is among the first to publicly test the technology, aiming to integrate AI systems that not only analyse markets but also can act on them.
While traditional AI tools have helped streamline data processing and flag investment signals, agentic AI goes a step further, enabling algorithms to simulate scenarios, adjust portfolios, and potentially execute trades, all within predefined limits. Robeco’s early experiments, combining LLMs with software functions, mark a shift in how asset managers approach portfolio construction, blending human oversight with machine autonomy in pursuit of greater efficiency and scale.
“It’s the next step beyond large language models,” said Mike Chen, head of a unit called ‘Next Generation Quant’ at Robeco. “LLMs can answer questions but can’t take action. Agentic AI adds that capability: executing trades, performing tasks, making decisions autonomously within boundaries.”
Chen and Kees Verbaas, global head of fundamental equity, spoke to Investment Officer about the firm’s experiments and vision for AI.
The firm’s internal experiments are still in development, yet they reflect a broader transformation underway in the investment industry. Asset managers, faced with tighter margins and growing data complexity, are beginning to see AI not just as an assistant but as a potential collaborator. At the centre of this shift lies a provocative question: if machines can learn, decide and act, what happens to the roles of analysts, portfolio managers and traders?
A new type of machine
Robeco’s reputation in quantitative investing stretches back decades. Its strategies have long relied on data-driven models grounded in academic research. But Chen’s Next Generation Quant initiative was designed to look beyond the established playbook. It operates as an experimental unit within the firm, tasked with exploring cutting-edge techniques and feeding successful ideas back into existing products or using them to build new ones.
“AI is an evolution of traditional quantitative methods. Traditional quant was prescriptive. AI is more adaptive,” he said. “It allows the algorithm to learn relationships instead of being told how to transform variables.”
“Eventually, we could have ‘analyst agents’ generating scenarios, ‘PM agents’ deciding impact on portfolios, and ‘trader agents’ executing, all overseen by a human.”
Mike Chen, Robeco.
This approach has already led to innovations like what Robeco calls the Dynamic Theme Machine, an AI-powered ETF strategy that detects emerging themes in global markets. The system identified topics such as digital restaurants during the pandemic and Formula One racing after the popularity of the Netflix series Drive to Survive, pointing investors to stocks such as Liberty Media and Ferrari. These are themes a human might recognise after the fact, but AI can catch them as they surface, drawing from vast amounts of unstructured data.
Now Robeco is taking the next step. Chen envisions a system in which specialised AI agents interact with each other to simulate and optimise investment decisions. One agent might analyse sentiment data. Another might model portfolio impacts. A third could execute orders in the market. Humans remain in charge, but the machines do much of the thinking, and most of the doing.
“Eventually, we could have ‘analyst agents’ generating scenarios, ‘PM agents’ deciding impact on portfolios, and ‘trader agents’ executing, all overseen by a human.”
Not in isolation
Robeco’s vision isn’t happening in isolation. A recent academic study at Cornell University explored what they call agentic finance, a new AI architecture that combines large language models with reinforcement learning to build autonomous investment agents. It’s a model that looks strikingly similar to what Chen describes: analyst agents feeding ideas, PM agents weighing the impact, and trader agents taking action, all in a loop overseen by humans.
The broader industry is also starting to take note. A 2024 report by PwC Switzerland described agentic AI as an “autonomous strategist,” capable of reshaping how investment firms operate. Rather than chasing trades or reacting to signals, these systems could soon be setting strategies in real time. The report recommends exactly the kind of sandbox experimentation Robeco is running, with a strong focus on explainability and control.
Human judgment still matters
Despite the growing sophistication of AI, Robeco executives emphasise that human oversight remains essential.
“Asset owners trust humans, not algorithms. We may delegate tasks, but humans remain orchestrators,” said Chen.
“Human judgment makes the difference,” Verbaas added. “How you interpret data, assess business models, project earnings. AI helps with efficiency, but the ultimate decisions rest with people.”
The efficiency boost is already noticeable. “Analysts can now write two to three investment cases a week instead of one. AI helps gather data quickly, allowing more time for judgment and insights,” he said.
Explainable AI
This kind of human-machine collaboration is redefining the relationship between technology and expertise. At Robeco, AI is not a black box. It is designed to be explainable, with clear reasoning behind its outputs.
“We’ve even turned machine learning into a ‘glass box’ through explainable AI,” said Chen, referring to the concept of ‘XAI’, which refers to methods and techniques used to make AI systems understandable to humans, particularly focusing on how decisions are made.
Culture over size
While some of the largest asset managers in the world are investing billions into AI infrastructure, Robeco believes its real advantage lies in culture. Innovation is embedded across its investment teams, not isolated in a lab, Chen and Verbaas explained. That mindset allows the firm to scale its ideas intelligently, testing new strategies with its own capital before introducing them to the market.
So, rather than rely on brute force or faster processors, the firm favours a more thoughtful approach.
“We’d much rather be clever,” said Chen. “For instance, instead of analysing what the CEO says, which is often biased, we might look at what the analysts say. Or even better, look at what’s not being said at all. It’s about applying similar technologies from a different angle. That kind of originality tends to provide a longer-lasting edge.”
‘Nozzle tweaks and better fins’
An internal incubator program helps Robeco test strategies with its own capital before going public. Alluding to the learning and innovation process at SpaceX, Chen added: “No explosions. Just nozzle tweaks and better fins, if you will.”
Robeco’s determined approach contrasts with moves elsewhere in the industry. Earlier this year, Dutch pension fund APG announced it was stepping away from quant investing, citing excessive complexity and unclear value. Chen sees it differently.
“Quant isn’t a black box. Every step is traceable and evidence-based. Complexity isn’t the same as mystery. To walk away now, amid this technological revolution, feels like missing a fundamental shift.”
“Fundamental investing is constantly evolving too. We’re upgrading our methods, supported by innovations from the quant side.”
Kees Verbaas, Robeco
Verbaas agrees. “Fundamental investing is constantly evolving too. We’re upgrading our methods, supported by innovations from the quant side.”
Investing in a different future
The rise of agentic AI is also changing what kind of talent asset managers seek. Financial knowledge remains critical, but firms now look for people who can bridge economics and engineering, who can build algorithms as well as understand them. Against that backdrop, the question rises if ‘quant’, a traditionally nerdy corner in asset management, should still be known as such now that AI is offering new roads into the future.
“Maybe,” said Chen. “It’s really technology-powered investing. But names aside, the essence is combining data, algorithms, and sound economic reasoning. There’s art and human judgment at the heart of it all, even with agentic AI.”
“There’s art and human judgment at the heart of it all, even with agentic AI.”
Mike Chen, Robeco.
As firms race to adopt these new tools, they face a deeper question: who will control the future of investing. And what skills will matter most? Will today’s CFA become tomorrow’s Python coder? Will the best portfolio managers also be AI strategists?
Robeco’s early experience with agentic AI makes clear that these are no longer hypothetical questions. In Rotterdam, the groundwork is already being laid.
“We’re right on the edge, just trying not to fall off. And it’s fun!” said Chen.