Particle.news

Goldman Warns AI Could Erode Bankers' Reasoning

Engineers are coding Goldman-specific 'skills' into internal AI agents to capture institutional knowledge and ensure outputs can be audited.

Overview

  • Chris Churchman, who spoke on Goldman’s podcast Monday, said outsourcing reasoning to AI risks 'cognitive atrophy' that could weaken bankers' ability to think from first principles.
  • Goldman has kept Marquee’s AI features internal while testing because finance demands consistently factual, traceable answers and the firm says it has not yet solved that challenge.
  • Engineers, led by CIO Marco Argenti, are building reusable 'skills' so AI agents act like knowledgeable Goldman insiders and developers are shifting from writing code to mentoring and auditing agents.
  • The firm has poured heavy resources into AI—reported spending was about $6 billion last year—and leaders warn the change could hollow out apprenticeship roles and reduce hands-on learning for junior staff.
  • Goldman’s own research estimated over 9% of U.S. jobs could be displaced by generative AI over a decade, highlighting both the firm’s push to capture returns and the wider workforce and governance trade-offs regulators and firms will face.