Will the AI Revolution Erode Human Expertise?
Updated: Oct 12
Over the past two weeks, I attended 2026 LSFI - Luxembourg Sustainable Finance Initiative summit and the LPEA - Luxembourg Private Equity & Venture Capital Association AI Lab at Linklaters . Both events featured insightful speakers sharing how their organisations are implementing AI and how they see this transformation reshaping the industry.
Across the industry, organisations are investing heavily in AI agents and redesigning their business models. The main drivers are clear: greater efficiency and lower costs. For now, human expertise remains at the centre of this transition. Professionals with 10 to 40 years of experience are teaching AI systems what good output looks like, applying judgement to their results, and building robust review and control processes around them. But a deeper concern is emerging AI may increasingly replace the tasks traditionally assigned to junior professionals. The next generation is learning to review AI-generated work through checklists, KPIs and error-control frameworks, yet many are no longer gaining the same hands-on experience. If learning by doing disappears, how will they develop the judgement needed to recognise when an output is merely compliant, but not genuinely correct?
Too often, juniors learn only the specific signals that suggest an AI output may be wrong and they do so under the supervision of a senior expert. The real challenge will come when those senior professionals "the people who built their expertise before AI" are no longer in-house. Who will be equipped to challenge, approve and take responsibility for AI-generated work with a truly expert perspective?
AI is evolving extraordinarily quickly. At some point, its outputs may become persuasive enough to pass our controls, or to make us believe they are correct, even when they are not.
I have seen a similar dynamic during fund-administration migration projects. I trained teams in Eastern Europe and Asia as NAV-calculation activities were transferred abroad, leaving mainly oversight and control responsibilities in Luxembourg. After 12 to 18 months, some new hires in Luxembourg struggled to respond directly to complex client requests. They could perform the prescribed checks, but without hands-on experience they often had to forward questions to the teams in Asia without fully understanding them. The knowledge had been transferred, but expertise remained with the people who practised the work every day, or with the subject-matter experts who had built that knowledge earlier.
This is the lesson I take from today’s AI transition: efficiency must not come at the cost of the learning pathways that create tomorrow’s experts. Organisations need to preserve opportunities for junior professionals to practise, question, make mistakes and understand the work beneath the output.
How is your organisation protecting human judgement while accelerating AI adoption?
Comments