Goldman Sachs: Junior Bankers Now Start Their Careers by Managing AI Agents

Kevin Sneader, president of Goldman Sachs for Asia-Pacific excluding Japan, said last week: "When our young employees come to work now, they manage [AI] agents." The line is short, but it comes from one of the world's largest investment banks, with about 46 thousand employees, where the largest and cheapest workforce is first-year analysts and associates. If their first task is now managing agents, the very ladder that people climb inside the bank is changing.

What we know and what we don't

Bloomberg quotes only the remark. The article does not say which agents Sneader means, which models they run on, how many people already work this way, or which tasks have been delegated to them. This is a statement from a top executive, not a deployment report, and it should be read as such.

Sneader himself joined Goldman in 2021 from McKinsey, where he was global managing partner. He is used to describing the transformation of large organizations in the language of management. So when he says "manage," it sounds like a job description rather than a metaphor.

How the bank got here

The public facts about Goldman known as of 2025 form a clear chain:

• January 2025: the corporate assistant GS AI Assistant launched as a pilot for roughly 10 thousand employees;
• June 2025: the tool was rolled out across the entire firm;
• July 2025: chief technology officer Marco Argenti described testing Devin from Cognition, an autonomous coding agent, and a "hybrid workforce" of people and agents;
• autumn 2025: as part of the OneGS 3.0 program, management announced a cap on headcount growth alongside expanded AI automation.

This adds context to Sneader's quote. The bank has steadily moved AI from a "chat assistant" mode to an "executor with its own assignment" mode. The takeaway for juniors is straightforward: a human sets the task, an agent does the draft work, and a human checks it.

Why the junior ranks came first

Traditional entry-level work at an investment bank consists of building financial models, preparing presentations, reconciling data and exports, and putting together comparison tables for companies. All of it is formalizable, repeatable and verifiable, which makes it almost an ideal fit for agents. Hence the logic: junior staff became the front line where automation is visible to the naked eye.

This is also where the main problem arises, and this is our assessment, not the bank's words. The profession of investment banking was long learned by hand: an instinct for errors came from hundreds of models built and documents proofread. If an agent writes the draft, the reviewer needs exactly that experience, and there is nowhere to get it. Then there is accountability: in a regulated industry, the final number is owned by a person with a name and a signature, not by a model.

What to do for those entering the profession and those hiring

For aspiring bankers and finance professionals, the practical takeaway is this: value is shifting from speed of execution to the quality of task-setting and checking. It is worth learning to phrase a task so that it can be verified, to keep a log of what the agent did and on what assumptions, and to recalculate selected results by hand so as not to lose the skill.

For managers and HR teams, the questions are different. How do you train people who do not go through the classic routine? How do you measure the performance of an "agent manager"? Who is responsible if an agent makes a mistake in a deal valuation model? Banks that answer these first will gain an advantage in hiring. The rest will have to catch up, and first they should demand from their own top executives not slogans but numbers: how many agents, on which tasks, and with what error rate.