The Real AI Risk in Investment Banking Isn’t Replacement - It’s Deformation
By Jurgis V. Oniunas, Chairman of IMAP. Adapted from remarks delivered at the IMAP global conference in Dublin, April 2026.
We have been considering the effects of AI on our industry for several years now, mostly in terms of efficiency and competitiveness - how to use it, and why we can’t afford not to use it. Earlier this year, the market had one of its periodic AI shock moments, sending software stocks down by twenty percent and more on the strength of a single AI model release, thus re-valuing a whole industry sector. We are now going through another one, sending the valuation of semiconductor producers, especially those concentrating on memory chips, through the roof. And we can expect a regular series of similar shocks as AI works its way through the world economy. We can also expect the continuation of the long-running debate about which jobs and how many jobs will be replaced by AI.
But the more important conversation, I think, is a different one, and it sits to the side of the replacement question.
AI, Hard Skills, and the Soft Skills Paradox
Here is the paradox: the more powerful our tools become, and the more extensively we use them, the more we need the soft skills of understanding that tell us what to do with these tools. And the clear danger is that the advancement of our soft skills is not keeping up the advancement of our new-found hard skills.
Hard skills are always attractive because they are visible. You can test them and you can train them. They produce outputs. They look efficient. They fit neatly into spreadsheets and performance reviews. Soft understanding, by contrast, is harder to measure, harder to certify, and harder to sell. When was the last time you really thought about your organization’s values, or presented them in a client pitch? There are plenty of memes on the internet that make fun of organizational values.
That something is hard to measure does not mean it is less important. In complex situations, it is usually the opposite. A person can have great technical skills and still completely misunderstand the situation. A company can have brilliant technology and still fail because of a broken culture. A government can have overwhelming military power and still lose the more complex struggle for legitimacy and for whatever objective it set out to achieve.
That is why an obsession with hard skills, and an unquestioning acceptance of all things AI, is dangerous. It mistakes performance for insight. AI is genuinely remarkable at pattern recognition, speed, and repetition. It can write, sort, summarize, predict, and generate at a scale no human can match. As it works its way through the economy, it may replace millions of jobs. It will probably, eventually, also create millions of jobs we cannot yet imagine.
Deformation, not Replacement: The Real AI Risk
The deeper concern should not be about replacement. It should be about deformation. The most important tasks in our business are not about processing information; they are about judgment, about trust, about timing, about knowing when a rule should be bent, when a metric might be misleading, and when a solution is technically right but essentially wrong. Those are the tasks that get hollowed out when the tools that surround them become too good at producing the appearance of an answer.
The same risk exists at the level of an institution. A firm can convince itself that better systems alone will produce better results. It installs new procedures, new dashboards, new software, new metrics. It removes layers of people. And it congratulates itself on becoming more efficient. But what if the core problem was never the process - and what if, by putting the process first, the firm is destroying its culture?
Culture is the invisible operating system of any organization. It embodies the values of everyone who works there, shapes who and what gets admired, decides who gets rewarded, promoted, defines what is tolerated, and governs how people deal with clients, with partners, and with each another. It is also the thing that AI cannot supply.
What the Thinkers Tell Us: Judgment, Power, and Complexity in the Age of AI
The tension between hard skills and soft skills is not new. Some of the most influential thinkers of the last century studied versions of it long before generative AI was a question.
Hannah Arendt argued in On Violence that power comes from people acting together - itself a profoundly soft-skill phenomenon - while violence is a coercive instrument that can destroy power but cannot create it. History keeps showing us that force alone cannot create legitimacy, trust, or a shared future. Violence is often a symptom of failure, not proof of strength.
James C. Scott, in Seeing Like a State, showed how large institutions prefer simplified versions of reality because they are easier to govern and manipulate. But human life is not simple. People live in local worlds shaped by memory, custom, adaptation, and practical knowledge. When outsiders override that complexity, they may act in ways that look rational on paper and fail in practice. That, incidentally, is one of the strengths of an partnership like ours - the ability to bridge those realities across six continents.
And Daniel Kahneman, in Thinking, Fast and Slow, showed how easily human judgment is distorted by shortcuts, overconfidence, and blind spots. We like to think that expertise guarantees good reasoning. It does not. People can be brilliant in narrow domains and systematically wrong in adjacent ones. A hard-skills culture tends to assume that competence in one area implies wisdom in general. It does not - many highly skilled people are trapped inside bad assumptions they never learned to see.
When I say, “soft skills,” I am not thinking about “people skills” in the casual sense. Those matter, but they are not the core. The deeper soft skills are judgment, empathy, historical memory, perspective, emotional intelligence, and the ability to understand whole systems rather than just fragments. These are the conditions of intelligent action in a complex world, and they are precisely what powerful instruments cannot supply on their own.
AI is giving us powerful instruments. But instruments need direction. They need purpose. They need interpretation. Without that, technical skill becomes a commodity. Management becomes bureaucracy. AI becomes automation without wisdom. Military power becomes force without politics. And progress becomes motion without meaning.
What This Means for How We Work at IMAP
This is the lens through which we think about the four values to which we are committed at IMAP - Executional Excellence, Creativity, Dedication, and Empathy. They are deliberately human, and at least for now, not completely subject to automation.
Executional Excellence is where AI shines most obviously: Beautiful presentations and brilliant financial models, at record speed. But pages of impressive charts can also conceal a flawed business model, a toxic working environment, or plain-old nonsense. Real executional excellence requires understanding the business, the trends in the market, and especially how AI itself is changing those trends.
Creativity is something that AI can support. It can flesh out alternative solutions to a given problem - but only if the inputs are themselves creative, and only if the outputs are filtered through hard-won experience and a sense of what is practically feasible.
Dedication is a mindset, not an automated to-do list. AI tools can keep us on track and help us use our time better, but dedication is the commitment to stick with a client through whatever it takes to reach the right outcome. Call it perseverance or grit. It is not a buzzword.
Empathy is the ability to understand the positions of clients and counterparties, and to work towards genuinely mutual solutions - and at the heart of every negotiation and every deal.
The Real Challenge
The future will not belong to the most technically proficient organizations. It will belong to those that can combine technical excellence with depth of interpretation - those that can pair technology with ethics, speed with historical memory, and power with understanding.
The real challenge is not whether we can do more. It is whether we know what should be done.
