Analysis
A Goldman Sachs partner publicly warned of a 'huge danger' in AI tools eroding junior bankers' underlying reasoning skills, CNBC reported, a notably candid admission from inside one of the earliest and heaviest institutional adopters of AI copilots in financial services.
The warning is specific in a way that distinguishes it from the broader jobs-displacement debate. The concern is not primarily that AI will replace analyst headcount, though that discussion continues separately -- it is that analysts who lean on AI to build models, draft memos and synthesize diligence materials may never develop the underlying judgment that used to come from doing that work manually, repeatedly, under time pressure, early in a career. That skill-formation process has historically been how investment banks trained the people who eventually became the partners making judgment calls on billion-dollar transactions.
“The warning is specific in a way that distinguishes it from the broader jobs-displacement debate.”
Investment banking adopted AI copilots aggressively over the past two years specifically because the work -- comparable company analysis, precedent transaction research, first-draft memo writing -- is exactly the kind of structured, document-heavy task large language models handle well. Every major bank has rolled out some version of AI tooling for junior analysts, marketed internally as freeing up time for higher-value judgment work rather than replacing that judgment.
The warning complicates that framing. If the repetitive grunt work was actually how junior bankers built the pattern recognition that becomes senior judgment, removing it doesn't free up time for judgment development -- it removes the mechanism by which judgment gets developed in the first place. That is a distinct failure mode from job displacement, and one that would not show up in headcount numbers or productivity metrics for years, only in the quality of decisions made by a cohort of managing directors a decade from now who spent their analyst years supervising AI output rather than building models themselves.
The same concern applies well beyond banking, to any profession where junior-level repetitive work has historically doubled as training. Law firms, consulting practices and engineering teams are all running some version of the same experiment simultaneously, largely without a clear answer yet.