Who Takes the Wheel? Reflections from the CFA Investment Industry Conference in Warsaw
A question for the room
Would you put 30% of your savings into a fund run entirely by AI with no human portfolio manager? Would your answer change if a human could override the AI's decisions?
Maciej Kowalczyk, Vice President of the Management Board at CFA Society Poland, opened our panel with that question at the 2026 CFA Investment Industry Conference in Warsaw on 5 November. He asked it again before we closed. Most people would hesitate before handing their car keys to software. The parallel with autonomous driving ran through the whole discussion.
Our COO and co-founder, Giacomo Barigazzi, shared the stage with Robert Süttinger, Managing Director at UBS Asset Management and Head of Asset Management in Poland, and Grzegorz Witkowski, CEO of Insignis TFI. Between them sat a global institution working within strict risk frameworks, an asset manager preparing to launch an AI-managed fund, and a fintech that builds the technology itself.
AI is already part of the process
The question is provocative, yet AI is already well-established in the industry. According to a Mercer survey published in February 2026, 55% of asset managers have integrated AI into at least one investment process, and 27% are running pilots or proofs of concept. Looking ahead, 91% expect to increase their use of AI over the next 12 months.
With adoption at that level, the conversation in Warsaw quickly moved to harder issues: how much of the decision AI should be trusted with, and who answers for the outcome when something goes wrong.
Six levels of autonomy
To give the debate a common language, the panel used a scale borrowed from the world of self-driving cars:
- No AI. People manage assets using deterministic models and systems.
- AI assists with summaries, drafts and research, while people decide and act.
- AI recommends, and a person approves every action.
- AI acts within hard limits, with a person monitoring in real time.
- AI acts on its own within a defined domain, and people handle the exceptions.
- AI sets and executes the strategy. People set the goals and hold the off switch.
Each panellist placed their organisation on the scale, making the differences among a global institution, a fund launcher and a technology provider easy to see. For most of our clients, the natural home today is level 2: the model does the heavy analytical lifting and the investment professional makes the call.
The Monday morning test
Picture a portfolio manager on a Monday morning. The market moved overnight; there are hundreds of possible positions, and only a few hours to decide. That is the moment Axyon IRIS was designed for.
The simplest way to describe IRIS is as an extra analyst on the team. Every day it reviews the entire investment universe, forecasts how each asset is likely to perform and ranks them with no manual intervention. This is the most autonomous thing AI does at Axyon AI and happens at a scale and frequency no team of analysts could match. What follows stays with people. Clients use the rankings for idea screening, to challenge a thesis or as a systematic input to discretionary decisions, while portfolio construction remains entirely in their hands.
As Giacomo pointed out, Axyon AI serves very global markets, so it was valuable to see how adoption, autonomy and impact are perceived at a local level. The underlying questions are the same everywhere. What changes from one market to the next is the pace and the comfort with letting a model take part in the decision. Wherever that comfort sits, our experience after ten years in live markets is that AI earns its place by strengthening human judgement and that trust grows when professionals can see how a model reaches its view.
Who is responsible when the model gets it wrong?
Returning to the autonomous car, the panel asked where responsibility lies when an AI-managed portfolio makes an error or simply underperforms. Is it the solution provider, the fund manager or the client? There is no tidy answer yet, which is precisely why explainability matters so much.
Portfolio managers must justify their decisions to investment committees, clients and regulators, and they will not act for long on signals they cannot understand. With IRIS Explainability, clients see the drivers behind each daily signal and how those drivers change over time.
Full explainability, however, is not a realistic goal. Deep learning models combine hundreds of features in ways no one can trace step by step. What professionals need is enough insight to understand a signal, question it and take responsibility for the decision that follows.
With thanks
We thank the CFA Society Poland team for the invitation and for hosting a day full of thoughtful discussions, and our fellow panellists for an exchange that brought very different perspectives to the same question.
As for that opening question, we will leave you to decide how you would vote.