This week, Citi told its investment banking analysts they'll make associate after two years instead of three. Private equity firms now recruit analysts in their first weeks on Wall Street, and Citi's co-head of North America investment banking called that "very unfortunate and to some extent disappointing." A faster promotion is Citi's way of getting them to stay.
It's being reported as a banking story, but law, accounting and consulting have the same problem. And AI is breaking the deal that used to pay for training juniors in the first place.
The grunt work used to pay for the apprenticeship
A first-year lawyer or auditor wasn't much use, but they could do the repetitive work: document review, reconciliations, comps tables, the fourth turn of the pitch book. Clients paid for it by the hour. That covered the junior's salary, and the learning came free on top. You picked up judgement by doing the boring work badly, over and over, with someone senior marking it up.
AI breaks that from both sides. The repetitive work is exactly what AI does well, so there's less of it to bill. And as I wrote in Edition #9, firms that bill by the hour pass those savings straight to the client. Once training stops paying for itself, it's just a cost, and one your competitors can poach. So most firms hire fewer juniors and recruit people someone else trained.
The bottom of the pyramid is shrinking
Stanford's Digital Economy Lab found that employment of 22 to 25 year olds in the most AI-exposed US jobs is now 19% below where it would be if it had kept pace with less-exposed peers, up from 15% a year earlier. Experienced workers show no gap, and the shortfall comes primarily from reduced hiring. Stanford doesn't call it proof that AI is the cause, though the pattern is hard to explain any other way.
Big Law hired 7.5% fewer entry-level associates from the class of 2025, according to NALP, the first decline since 2014. A leaked PwC US deck reported by Business Insider showed plans to cut entry-level hiring by about a third over three years (PwC didn't confirm the numbers). Jamie Dimon said in May that JPMorgan will hire "more AI people and fewer bankers in certain categories," and last week Accenture said hiring next year will run "at a lower rate in part due to AI."
Not everyone is cutting
McKinsey plans to grow its North American workforce by 12% in 2026, including more entry-level hires. Latham & Watkins is growing its US summer associate class from 122 to around 170.
I think pricing power explains the split. Between a quarter and a third of McKinsey's revenue now comes from outcome-based work, as I covered in Edition #13, so it keeps the savings when AI speeds that work up. Latham still bills by the hour, but at rates few firms can match, so it can afford to treat training as an investment. Mid-market firms billing by the hour for more commoditised work can do neither. They cut juniors and hire experienced people trained by the firms that could still afford to.
The free-rider problem, AI-natives included
The firms that pay for training keep losing what they paid for. Private equity recruits analysts that banks trained. AI companies hired 46 lawyers from Am Law 200 firms in the first half of this year, and 40 of them were associates. Harvey took 22 while selling its software to 80% of the Am Law 100.
AI-native firms rely on this too. Falcon, a new Belgian AI-native law firm advised by former A&O global senior partner Wim Dejonghe, has every piece of work signed off by a qualified lawyer. Most AI-native services firms work the same way: AI does the production work and a qualified professional signs. That professional was usually trained inside the pyramid the AI-native firm is competing against.
In Edition #12 I argued that a services firm's lasting value sits with its licence-holders... the people who can sign the audit or the opinion. If incumbents train fewer of them and AI-natives produce almost none, signatures get scarce and their price goes up.
Can't AI train them too?
The obvious objection is that AI can fill the gap. AI tutors and simulations give juniors instant feedback, and in theory they could compress years of learning into months.
I think that's partly right. AI is a good teacher for the technical side, like building a model or drafting a clause. Judgement is harder. You build it by making calls that matter and having someone senior tell you when you got one wrong, often in front of a client who noticed. I haven't yet seen a simulation that recreates that pressure.
What a new apprenticeship could look like
If the grunt work no longer trains people, firms have to design the training on purpose. A few have started.
KPMG UK now front-loads exams and pairs graduates with experienced staff before they meet clients. Its chief people officer says graduates will spend more time "reviewing and exercising judgment rather than simply producing outputs." The junior's job moves from writing the first draft to checking the AI's, which is exactly where judgement gets built. Ropes & Gray takes a different route and buys the time back, letting first-year associates put up to 400 of their 1,900 billable hours towards learning AI tools.
While juniors are billed by the hour, every hour spent learning is an hour the client won't pay for. On fixed-fee or outcome-based work, a firm can staff a junior next to a senior, accept a slightly lower margin on that job, and treat the difference as an investment. I think the firms that move fastest to fixed fees will be the ones that can still afford to train.
What I'll be watching
1. A licence-holder squeeze around 2030. The juniors firms aren't hiring today would have been the qualified mid-levels of 2030. I believe lateral pay for qualified lawyers and auditors jumps, and revenue per licence-holder becomes the number buyers care about most.
2. Banking's retention rules spread. JPMorgan has threatened to fire analysts who accept future private equity jobs in their first 18 months, and Goldman has asked juniors to certify they haven't signed elsewhere. I expect a major law or accounting firm to try something similar within a year.
3. A Big Four firm follows Ropes & Gray. My feeling is that within 12 months, a Big Four firm carves out protected training time for juniors on the same scale as Ropes & Gray's 400 hours. The firms that do it will own the partners of 2035, and the rest will be hiring them.
If you run a firm... are you still training juniors, or waiting to hire them once someone else has?
Hit reply, I read everything.
See you next time,
Nikola