Goldman Sachs Says New Hires Will Manage AI Agents From Day One. Who Teaches Them Judgment?
A Goldman Sachs executive said in October 2026 that the bank's newest hires now manage AI agents from the start, and that middle managers face a 'generational challenge.' The junior work that trained every banker is moving to software. What that means for how small teams train new people.
Quick answer
On 8 October 2026 Goldman Sachs Asia-Pacific head Kevin Sneader said the bank's newest hires now manage AI agents from the start of their careers, Bloomberg reported. The routine work that used to train junior bankers is moving to software, which raises a practical question for every employer: how new people learn to spot a wrong answer.
Goldman Sachs says its newest hires now manage AI agents from their first day. Kevin Sneader, who runs the bank's Asia-Pacific business, said it on 8 October 2026, and added that middle managers may face a "generational challenge." The routine work that trained every junior banker is moving to software. The open question is how new people learn judgment without it.
What Goldman said
Sneader spoke at the Milken Institute Asia Summit in Singapore. Bloomberg reported his line: "When our young people start now, they manage agents."
Coverage of the remarks adds context on how far the bank has gone:
- Anthropic's Claude is deployed in trade accounting, compliance and client onboarding.
- Goldman reports productivity gains of more than 20% in software engineering tasks since introducing agentic AI.
- Chief information officer Marco Argenti has said junior staff need to get good at describing tasks, handing them off and supervising AI systems.
Goldman has not announced junior hiring cuts tied to this. In 2025, CEO David Solomon said he did not expect the bank to simply end up with fewer people. The change being described is to what juniors do, not yet how many there are.
The apprenticeship that is disappearing
For decades, the first two years in investment banking were an apprenticeship disguised as grunt work. Analysts built financial models, formatted presentations and checked numbers late into the night.
The work was repetitive and often miserable. It was also how people learned what a sound deal looks like, where models break and which numbers deserve a second look.
That repetitive layer is exactly what agents handle best. Remove it and you remove the practice that built judgment.
Who checks the agents?
Sneader's two remarks sit together awkwardly. New hires supervise agents. Middle managers, who used to review the juniors' work, face a "generational challenge."
Supervising an agent means catching its mistakes. You can only catch a mistake you can recognize. If the people at the bottom never did the work by hand, and the layer that used to check them is thinning, the review chain gets weaker at both ends.
What this means for a small business
Small firms are running a quieter version of this. The new hire who used to learn by doing the bookkeeping, the first drafts or the data entry is now handed a tool that does it. That is faster. It also means nobody new is building the instinct that says "this number looks wrong."
A practical approach:
- Keep the reps. Have new people do core tasks by hand for a set period, even when software could do them, until they know what good looks like.
- Then hand them the agent. Their first job with the tool is reviewing its output against the standard they learned.
- Name a checker. Every reviewer needs someone experienced who spot-checks their reviews, at least at first.
- Write down the failure modes. Each time the tool gets something wrong, record how it was caught. That list becomes your training material.
A team that only supervises software it does not understand is not really supervising it.
The question to ask this week
How does a new person in your business learn to spot a wrong answer, if the software is now doing the work they used to learn on?
Sources
- Bloomberg, "AI in Banking: Junior Bankers Find Work Managing AI Agents," 12 October 2026 (Sneader quote as reported).
- Crypto Briefing, "Goldman Sachs' new hires will manage AI agents from day one," 8 October 2026.
- Axios, "AI isn't taking banking jobs, Goldman Sachs CEO says," October 2025.
Frequently asked questions
- What did Goldman Sachs say about new hires and AI agents?
- Kevin Sneader, who runs Goldman Sachs in Asia-Pacific, said at the Milken Institute Asia Summit in Singapore on 8 October 2026 that new hires now manage AI agents from day one. Bloomberg reported his line as 'When our young people start now, they manage agents.' He also said middle managers may face a 'generational challenge.'
- How is Goldman Sachs using AI agents?
- Coverage of the remarks reports Goldman has deployed Anthropic's Claude in trade accounting, compliance and client onboarding, and reports productivity gains of more than 20% in software engineering tasks since introducing agentic AI.
- Will AI replace entry-level jobs?
- It is changing them first. The routine tasks juniors used to learn on, such as building models, formatting documents and first drafts, are what agents handle best. Employers that still hire juniors need a new way to build their judgment.
- How should a small business train new staff when AI does the routine work?
- Keep deliberate practice: have new people do key tasks by hand often enough to know what a good result looks like, review agent output against that standard, and give them a named person to check their reviews.