What happens to the junior lawyer pipeline after AI

Every firm I talk to has a view on what AI will do to its margins. Far fewer have a view on what it will do to its juniors, and the ones that do tend to hold a comfortable one: trainees will simply spend less time on drudgery and more time on interesting work. I’d love that to be true. But it rests on an assumption I don’t think holds, which is that the drudgery was only ever drudgery. A lot of it was training. We just never called it that, because it happened to be billable at the same time.

What the grind was really teaching

When I was producing reports on title as a junior real-estate lawyer, I didn’t experience it as education. It felt like repetition. But somewhere around the hundredth one, something changed. I started to know what normal looked like - the usual covenants, the standard rights of way, the typical shape of a clean title - so that the odd entry jumped out before I’d consciously read it. That instinct for the unusual sits underneath almost every senior judgement in practice, and I acquired it the only way anyone did: by seeing the same thing over and over until the exceptions became visible.

Nobody designed that curriculum. It was a by-product of the work being done by hand. Take away the hand-done work and you take away the curriculum with it, whether you meant to or not.

Reviewing is not the same as doing

The usual answer is that juniors will now review the machine’s output instead, and learn that way. Partly, they will. But reviewing something you’ve never built is a different skill, and a much weaker teacher. If you’ve drafted fifty leases, you know where they go wrong because you’ve made those mistakes yourself. If you’ve only ever checked AI drafts, you know what a plausible lease looks like, which is not the same thing at all.

There’s also a familiar human tendency to trust automated output more as it becomes more reliable. A junior checking a tool that is right nineteen times out of twenty will, quite reasonably, start to skim. The twentieth time is the one that matters, and it’s exactly the one they’ve been conditioned not to catch.

The gap shows up late

The cruel thing about this problem is its timing. Nothing looks wrong this year. The work gets done, the clients are happy, the margins look better. The cost arrives five or eight years later, as a cohort of senior associates who have run plenty of matters but have a thinner sense of what can go wrong, and a smaller pool of people genuinely ready for partnership. By the time it’s visible, the years that should have built those people have already passed.

That’s why this can’t be left to sort itself out. It has to be designed, deliberately, and it has to be designed now.

Do some of it by hand, on purpose

Other professions that automated their routine work have faced a version of this before us. Aviation has long worried about pilots who rarely fly manually and are then expected to take over when the automation hands back control. The answer there wasn’t to remove the automation. It was to build deliberate manual practice back into training.

Law can do the same. Have trainees do a defined set of tasks by hand early on - not because it’s efficient, but because it’s how the pattern library gets built. Use the tool’s real mistakes as teaching material: a bank of AI outputs with known errors, which juniors have to find before they’re trusted to review live work. Make “here’s what I’d have done differently, and why” part of every review, so checking becomes an active exercise rather than a passive one.

Make supervision an actual job

Most of what juniors learned from seniors used to happen incidentally - a comment in the margin, a call overheard, a mark-up returned at 7pm. With less routine work flowing up and down the chain, those moments become rarer. Supervision has to stop being a side effect of the work and become part of it: time that is protected, recognised and, frankly, counted in how seniors are assessed. If a partner’s development of juniors never shows up anywhere that matters to their career, it will be the first thing to go when a deal gets busy.

Hire for checking, not just stamina

Finally, the traits that make a good junior are shifting. Stamina and accuracy over long hours mattered when the job was volume. Now it matters more that a junior is sceptical by temperament, curious about why an answer is right, and comfortable saying “this doesn’t look right” to a tool and to a supervisor. Those qualities can be screened for and developed, but only if firms decide that’s what they’re looking for, and say so in how they recruit and how they review.

AI doesn’t end the junior lawyer. It ends the accidental apprenticeship - the one that worked because the boring work happened to teach you everything. The firms that recognise that and build a deliberate apprenticeship in its place will have the best senior lawyers in a decade. The firms that bank the efficiency and assume the pipeline will look after itself will discover, too late, that they automated away the very thing that was making their partners.