Goldman Sachs is pushing artificial intelligence deeper into institutional finance, but one of the executives building that technology is warning that automation could undermine the reasoning skills Wall Street relies on to develop its next generation of bankers and traders.
Chris Churchman, a Goldman Sachs partner who leads its Marquee digital platform for institutional clients, warned of a “huge danger” that workers could outsource their reasoning to AI models, creating what he called “cognitive atrophy” that prevents people from reasoning from first principles.
HRD America similarly reports that Churchman sees the “real danger” as diminishing reasoning capacity, arguing that technology has historically caused skills to decline once machines become better at performing them.
AI Efficiency Creates an Apprenticeship Problem
The concern is particularly relevant to investment banking because junior employees traditionally learn through repetitive analytical work that AI can increasingly automate.
CNBC reports that Churchman believes banks need to balance AI adoption with Wall Street’s apprenticeship culture. Junior traders, for example, develop judgment by interpreting client pricing requests, using pricing tools and discussing decisions with experienced risk takers—tasks that can now increasingly be automated.
Churchman’s concern is not simply that jobs disappear. The deeper risk is that employees stop performing the work through which expertise is created.
HRD America quotes him saying businesses should make a “proactive choice” to empower human reasoning rather than delegate it. He wants future Goldman employees to retain unusually strong reasoning capabilities even after years of working alongside AI.
Goldman’s AI Still Faces a Reliability Problem
The warning comes from an executive directly involved in building AI products rather than from someone arguing against the technology.
Churchman runs Marquee, Goldman’s platform providing institutional clients with market data, research, risk analytics and trade execution services.
CNBC reports that the bank is developing AI capabilities around the platform, although its Marquee AI system is currently available only to Goldman employees.
Its biggest technical challenge is accuracy. Churchman said institutional AI needs outputs that are “100% factual” and auditable, because financial institutions cannot tolerate hallucinations in the same way a consumer chatbot might.
That creates an important product constraint: Goldman wants AI capable of doing more analytical work while simultaneously ensuring humans remain capable of identifying when that analysis is wrong.
The Bigger AI Product Question Is What Humans Still Need to Practice
HRD America reports broader anxiety about this trade-off, citing workforce research in which 66% of employees worried AI would erode their writing skills and creativity, while 63% expressed concern about memory and 60% about social skills.
For founders and product leaders, Goldman’s dilemma points to a larger design question.
Most enterprise AI products are optimized around how much work they can remove. But in professions built around judgment—finance, law, engineering or medicine—some inefficient work also functions as training infrastructure.
AI may produce the spreadsheet, analysis or recommendation faster. The difficult product decision is deciding which tasks should remain human long enough for the next generation to learn how to challenge the machine.
That could make one of AI’s most valuable enterprise features surprisingly counterintuitive: knowing when not to automate the reasoning itself.
