How AI Shrank Junior Investment Banking Jobs Since 2024

An IPO prospectus. The S-one. Six people, two weeks, nights and weekends. David Solomon, chief executive of Goldman Sachs, on stage at the Cisco AI Summit in January twenty twenty-five, says AI can draft ninety-five percent of it. In minutes.

Ninety-five percent? That's not a productivity tweak. That's the entire job of a first-year analyst.

Which is why I want to ask a question that sounds small and isn't. If the machine does the bottom rung of the ladder... where does Wall Street get its managing directors in ten years?

Okay. Back up. Why did banks ever want armies of twenty-two-year-olds?

The leverage model. One managing director wins the mandate, and under him there's a pyramid. Associates, analysts. The analyst is a fixed cost. Salaried, no billable hour meter, so the bank extracts as many hours as a human body allows. Eighty to a hundred a week, by Fortune's reporting. Comps tables, pitchbooks, formatting slide forty of forty.

Miserable. But also... that's the training, isn't it.

It IS the training. You build four hundred comps tables, and somewhere around table three hundred you start to feel when a number looks wrong. Nobody teaches that in a classroom.

So what walked in and took that work?

JPMorgan's internal tool. LLM Suite, released summer twenty twenty-four. Two hundred thousand users in eight months. Today around two hundred and fifty thousand employees have access, basically the whole firm minus branch and call centre staff, and roughly half of them use it every day.

That's not a pilot. That's plumbing.

And listen to how JPMorgan's John Waldron described one deck, put together on Nvidia, speaking to CNBC. He said, you can imagine in the past how that would have been done. They would have had teams of investment banking analysts working long hours at night to do it.

Past tense. He's talking about those analysts in the past tense.

Goldman has its own assistant for bankers. Morgan Stanley put one in front of nearly all its advisor teams. And Goldman's own researchers reckon AI could automate about a quarter of current banking work hours. Modelling, note-taking, spreadsheets, deck formatting. The junior stuff. Most exposed of all.

Sure, but a quarter of hours isn't a quarter of people. Is anyone actually hiring fewer humans?

Here's the number that stopped me. Debasish Patnaik, who runs McKinsey's AI arm QuantumBlack, told Bloomberg that banks are cutting junior analyst classes by as much as two thirds.

Two THIRDS?

At some institutions. The New York Times reported firms were weighing that kind of pullback back in twenty twenty-four, Goldman and Morgan Stanley among them. And the bosses aren't hiding it. Jamie Dimon: AI will eliminate jobs. Jane Fraser at Citi: some roles will no longer be required. Société Générale's chief executive, on costs... nothing is sacred.

Hang on though. Banks cut juniors every time deals dry up. That's a cycle, not a robot.

Strongest counterargument there is, and it deserves room. Jason Napier, who runs European bank research at UBS, said flatly that banks aren't delivering improved efficiency yet. Cost bases are large, the tools aren't fully implemented. And Goldman's headcount didn't fall. About forty-eight thousand three hundred last September. Up roughly eighteen hundred on the year.

So AI might just be a flattering label on ordinary cost discipline.

Partly. Solomon says he's not in the job apocalypse camp. He told Fortune that if they get this right, he doesn't think it significantly lowers the number of people they have. Then he added that AI means more valuable people doing more valuable things.

Nobody's ever been laid off by the phrase more valuable.

But underneath the cycle argument there's a pattern. The Canaries in the Coal Mine study, built on payroll data from millions of American workers. In the most AI-exposed occupations, employment for twenty-two to twenty-five year olds FELL about six percent between late twenty twenty-two and July twenty twenty-five. Workers thirty-five to forty-nine, same jobs... grew over nine percent.

Same industry. Opposite directions. Purely by age.

And pay isn't what's moving. Salaries held. It's the number of doors.

Then what's left for a human analyst?

Solomon's last five percent. Judgement. Knowing which comparable company actually belongs in the set. Sitting in the room when a founder is lying to himself about valuation. Being accountable when the number's wrong, because you can't fire a model. Although... juniors told Fortune they're still buried in grunt work the AI was supposed to take.

Of course they are. Okay, so back to your question. Where do the MDs come from?

Nobody has an answer. Patnaik put it like this: banking is an apprenticeship business, today's junior analysts become tomorrow's managing directors. And then the line I can't stop thinking about. Senior judgement cannot be manufactured laterally.

You can't hire experience nobody was allowed to accumulate.

And here's the irony. Those same junior cohorts supply roughly sixty-two percent of the AI talent the banks are recruiting. The people being thinned out are the people who understand the machine best. A JPMorgan executive told the Financial Times that if junior bankers never learn the fundamentals, it could come back to haunt the whole industry.

So the saving lands this quarter. The bill arrives in twenty thirty-five.

That's the trade. And it isn't only banking. Law, accounting, journalism. Any job where the grunt work was secretly the apprenticeship is running this experiment on itself right now. The machine absorbed the boring part. Turns out the boring part was the classroom.

And the exam results come back in a decade.

Wall Street built a ladder out of tedium. Then it automated the bottom rungs... and left a generation standing underneath it, looking up.