Goldman Sachs exec warns AI could quietly hollow out the next generation of Wall Street talent
A senior Goldman partner says banks risk producing workers who can't think for themselves if AI handles all the analytical work that used to train junior staff.
Key points
- Goldman Sachs partner Chris Churchman warned in May 2025 that over-relying on AI risks causing "cognitive atrophy" in the next generation of bankers.
- Churchman leads Marquee, Goldman's digital trading platform used by hedge funds and other large institutional clients.
- Goldman's internal AI version of Marquee is currently restricted to Goldman employees and has not been released to clients.
- Churchman said even Goldman has not yet "figured out" how to manage the shift to AI-assisted finance.
- The AI system Goldman tested admitted it was "better at sounding thorough than being thorough."
A Goldman Sachs partner has sounded a sharp alarm about what happens when banks hand too much thinking over to artificial intelligence, software that can answer questions and analyse data at machine speed.
Chris Churchman, who runs a digital trading platform at Goldman called Marquee, put it plainly: "There's a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves." Cognitive atrophy means the gradual weakening of a mental skill through lack of use, the same way a muscle shrinks if you stop exercising it.
The remarks came from a transcript of Goldman's internal "Exchanges" podcast, first reported by CNBC Tech, as banks race to weave AI into trading and client services.
Why does this matter for ordinary workers?
It isn't just a Wall Street problem. The same pattern Churchman describes could repeat in any field where AI takes over the entry-level work that teaches beginners how to think.
At banks today, junior traders learn by taking live client orders under the eye of a senior colleague. Make a mistake, get corrected, gradually build judgement. Churchman's worry is direct: if AI handles those requests automatically, the junior trader never makes the mistakes, never builds the judgement.
"You learn by doing, and a lot of knowledge is tacit, it was never written down," he said. Tacit knowledge is the kind of know-how that lives in a person's instincts, the feeling an experienced trader gets about a market, the thing that is almost impossible to put in a manual.
Lose that handover between generations, and firms could end up with senior staff who understand everything and junior staff who understand almost nothing. Our earlier coverage of how banks are ceding AI decisions to a handful of tech giants flagged a related risk: the finance industry is building critical dependence before it has fully thought through the consequences.
Can banks have AI and still train people properly?
Churchman believes yes, but says Goldman hasn't worked out how yet. That admission from one of the world's most profitable investment banks is striking.
His prescription: AI should handle repetitive, low-stakes tasks while humans keep real authority over high-stakes, uncertain calls. Systems must be built so employees stay in the driving seat rather than becoming passive button-pressers who rubber-stamp whatever the machine suggests.
The honesty challenge is just as thorny. Consumer chatbots carry polite warnings that they can make mistakes. In finance, a wrong number in a risk report can cost millions. When Goldman tested its Marquee AI hard, the system gave an answer Churchman found both unsettling and oddly candid: "In the end, I'm better at sounding thorough than being thorough."
Closing that gap between confident-sounding output and genuinely accurate output is the central technical problem Goldman is still trying to solve before the platform opens to clients.
What happens next?
Marquee's AI tools remain Goldman-only for now. Churchman, who also co-chairs Goldman's internal AI working group for its banking and markets division, says the firm needs to design workflows that force employees to reason through problems rather than simply accept AI answers.
The real watch point here isn't whether Goldman eventually ships the product. It's whether the apprenticeship model that produces senior talent survives long enough to matter. Push back on the machine, check its reasoning, and make sure you could still do the job if the software disappeared tomorrow. That advice applies well beyond Wall Street.
