When it comes to the empirical implementation of AI, what should trading desks bear in mind?
If we look back at the previous generation of AI models, their success largely depended on the availability of large and relevant datasets for training. With the emergence of Generative AI, a new component has entered the equation: the user. In addition to the model and its data, the human interaction layer now plays a decisive role in shaping outcomes.
The main challenge of Generative AI is therefore less technological than cultural. Governance, accessibility, and integration are key. Firms should focus on embedding the technology into existing workflows such as internal messaging systems or portfolio management platforms, so that users can engage with it naturally. At H2O AM, we learned that it’s better to start small, with simple, verifiable tasks, and gradually evolve from productivity gains to actual innovation. For instance, instead of ingesting the largest possible dataset of research documents, we asked our portfolio managers to manually curate what truly mattered. The result was a smaller but far more relevant dataset which also created a stronger sense of ownership and engagement from users.
How helpful are discussions around the theoretical use cases for AI in trading – is the sky really the limit?
There is no shortage of speculation about what AI might one day accomplish, from agentic models to fully autonomous trading, but the real transformation is already underway. Today’s large language models are more than capable of reshaping how we work.
In practice, each new and more powerful model release doesn’t necessarily solve the fundamental challenge of integration. What matters is not the model’s size or complexity, but how intelligently professionals use it within their workflows. The key question has shifted from “what can AI do?” to “how do we use it purposefully?”. The real opportunity lies not in dreaming about the limitless future, but in mastering the practical applications available right now.
Looking at the current state of play, how are LLMs and ML driving traders’ strategies today?
The influence of AI varies by investment style, but for discretionary managers like us, Generative AI is becoming an integral part of our existing investment process rather than a replacement for it. It provides us with a new set of tools to observe both ourselves and the market, acting as what we call a mirror and a sensor..
As a mirror, it supports our work in behavioural finance which remains a cornerstone of our discretionary approach. By analysing transcripts of our internal meetings, AI helps identify cognitive and group biases such as confirmation or overconfidence, offering an objective lens on our decision-making. As a sensor, it processes vast amounts of unstructured data, summarising external research, identifying consensus, and highlighting contrarian insights. Together, these applications sharpen judgement and deepen market understanding without eroding the human element.
From your perspective, what market structure change will have the biggest impact on the buy side going forward – AI, or something else?
The most significant change ahead for the buy side could be the emergence of what we call Generative Finance. Instead of investing in AI, for instance through exposure to technology stocks, a new management style will aim at investing with AI, embedding it directly within the investment process. This evolution will blur the divide between quantitative and qualitative analysis, enabling managers to combine human insight with machine reasoning more effectively.
Generative Finance could mark a shift from automation to augmentation, helping discretionary portfolio managers to think faster, see broader, and decide more rationally.