For most of the last century, nobody had to design a system for turning law graduates into lawyers. The work did it. A new lawyer spent years reviewing documents, checking citations, marking up agreements, and writing first drafts that someone more senior took apart. Clients paid for those hours. Judgment was the by-product.
Software can now do a good part of that work in minutes. The usual question is whether AI will replace lawyers. The more useful one is narrower. If the machine does the junior work, how do we make senior lawyers?
The old bargain was an accident
The traditional firm is a pyramid: many juniors, fewer seniors, and client work that flows down to whoever can do it most cheaply. That structure paid for two things at once. It produced the firm's profit, and it produced the firm's next generation, because the volume of work was also the curriculum.
Stanford Law School ran a course this fall on the future of the law firm. Professor David Freeman Engstrom, who led it, put the training problem in one line: "Young lawyers have always learned judgment through volume: document review, due diligence, the fiftieth markup." The question, he said, is which forms of repetition and learning by doing are genuinely developmental. One of the course's lecturers, Rachel George, added that much of what is known about where firms are heading comes from deal announcements, press coverage, and practitioner commentary. What follows is an argument from practice, not a finding.
Not all of it trained anyone
It helps to sort the old junior workload into three piles.
Work that builds judgment. A first draft of an argument. A chronology built from the raw record. An outline for examining a witness. Each forces the junior to decide what matters and to be wrong about it in front of someone who knows better.
Work that is only repetitive. Formatting, assembling records, conforming citations, and numbering exhibits. Nobody became a better lawyer the fortieth time. The machine is welcome to it.
The awkward middle. Document review is the best example. It is tedious, and it is also how a junior learns the case, the client's business, and how people write when they do not expect a lawyer to read it. A tool that ranks and summarizes the documents delivers the answer without the portable familiarity. The lawyer who knows the documents is the one who can cross-examine on them.
The mistake is to treat the third pile like the second.
Reviewing is not the same as producing
The model now emerging is that the machine drafts and the junior reviews. That model assumes that reviewing is the easier job. Unfortunately, it is the harder job.
The courts have already seen what happens when the reviewer cannot tell good output from bad. In Zhang v Chen, 2024 BCSC 285, a lawyer filed a notice of application citing two cases that ChatGPT had invented. The court accepted that she had not meant to deceive anyone, and ordered her to pay personally the costs her mistake had caused (at paras 32, 43). She held several law degrees, including a PhD in law (at para 27). Justice Masuhara observed that "there is a significant difference between academics and lawyering" (at para 32). He also cited research finding that large language models “cannot always predict, or do not always know, when they are producing legal hallucinations“ (at para 38).
Knowing the law in the abstract was not enough. What was missing was the practised habit of doubting a fluent answer—and going to the source. That habit is a senior skill, and it comes from having produced the work yourself, badly at first. As the court put it, "generative AI is still no substitute for the professional expertise that the justice system requires of lawyers" (at para 46). A junior cannot skip to editing.
The pressure runs the wrong way
In principle, the time AI saves could go to mentoring and reflection. In practice, clients expect to see the saving on the invoice, and firms are inclined to fill the freed hours with more files.
Training was never a line item. It was hidden inside billable work. When that work disappears, the training has to be paid for deliberately, and the firm is the one that pays.
That is an investment, and an exposed one. As we wrote about the collapse of Weil, a law firm's value is its people, and its people can leave. Large firms can respond by hiring experienced lawyers from elsewhere. But the lateral market only moves senior lawyers around. Someone, somehow, still has to make them.
What a small firm can and cannot do
A small firm has two advantages here. Its economics do not depend on a wide base of junior hours, so it has less reason to keep juniors on work a machine should do. And its juniors are close to the decisions. They are in the strategy call, the client meeting, and the cross-examination, and they can see why a choice was made as well as what it was.
It also has real disadvantages. There is less volume, so fewer repetitions. There is no formal program. Everything depends on a senior lawyer's willingness to stop and explain.
A few practices follow from all of this, for a firm of any size:
By hand first. A junior does a task unaided once or twice before using the tool for it, so that they know what a good result looks like.
Check against the source. AI output is verified against the statute, the case, or the document, and never against the reviewer's sense of what sounds right.
Explain the edits. A marked-up draft teaches little. Ten minutes on why the changes were made teaches a great deal.
Whole things, not slices. A junior who runs a small motion from start to finish learns more than one who drafts a section of a large one.
Why firms are still hiring
Senior lawyers are made slowly, by doing things poorly under supervision until they do them well. The machine has taken away part of the work. It has not taken away the need.
The firms that work out which part of the old drudgery was the education, and keep that part on purpose, will have senior lawyers in a decade. The rest will be bidding for them.
This post is general commentary. It is not legal advice about your situation.

