The Hidden Economics of the Vanishing Apprenticeship

Two columns ago, I followed the money through a law firm’s student recruitment program and found nearly $500,000 a year in lawyer time and expenses before a single billable hour was worked. Last column, I followed that same money into the delegation of actual work and found that firms were quietly losing much of it back through vague instructions, overworked memos, and write-downs nobody quite wants to talk about.

This column follows that same money to where it gets spent – on the reps junior lawyers used to log before anyone trusted them with something that mattered. Those reps are what’s disappearing now, and firms need to decide who is responsible for replacing them.

The job that built judgment

For as long as I’ve been watching firms operate from the management side, the deal was straightforward, if rarely said out loud. Juniors did the unglamorous work including the first-pass research memo, the lease-by-lease due diligence review, and the first draft of a statement of claim nobody expected to survive a partner’s red pen. It was slow. It was often duplicative. And it was also, not incidentally, how judgment got built.

Paul Saunders, chief strategy and innovation officer at Stewart McKelvey in Halifax, has been blunt about what’s changing. He learned contract law, he says, by grinding through leases line by line on due diligence files — the kind of task that now takes a junior minutes to summarize with an AI tool instead of days to read. He calls the resulting gap the “AI training conundrum,” and the phrase has stuck for a reason: it names a problem firms haven’t solved, not one they’ve avoided.

He’s not alone in naming it. Matthew Peters, national leader of transformation at McCarthy Tétrault, describes lawyers regularly stunned by how fast AI now compresses work that used to take hours. Erin Cowling, CEO of Flex Legal Network, has watched smaller, tech-forward firms shift first-draft pleadings — traditionally a new call’s job — onto AI entirely. And Allison Speigel, a partner at Speigel Nichols Fox in Mississauga, told Canadian Lawyer that even when an AI summary is “totally correct,” she still finds herself wanting to go to the underlying documents, because in litigation a few words can change a case.

That instinct isn’t just caution. It’s the job. The Law Society of Ontario’s guidance on lawyers’ use of AI is explicit that AI-generated output must be independently verified by a human, not by the tool that produced it. The Law Society of British Columbia has taken a similarly firm stance since 2023 through its own practice resource on generative AI, warning that the technology is no substitute for a lawyer’s professional judgment. Which raises the obvious follow-up question: verified by which human, exactly, if the humans coming up through the system never did the underlying work themselves?

Is “apprenticeship” still the right word?

I’ve gone back and forth on that word. It’s the one Axios, Bloomberg Law, and Canadian Lawyer have all used this year to describe this exact shift. But it also carries a certain nostalgia: structured, intentional, a master craftsperson patiently transferring a trade. Anyone who has actually been a junior associate knows that’s a generous description of what articling and the first few call years have often looked like: inconsistent supervision, wildly uneven feedback, and, as I wrote last column, a fair amount of student time quietly written off because nobody explained the assignment properly in the first place.

So no, I don’t think we’re losing some golden age of mentorship. I think we’re losing the reps. The volume of low-stakes, moderately tedious work that let a junior lawyer be wrong in private, a few hundred times, before they had to be right in front of a client. That’s the specific thing worth protecting. Not the word, the reps behind it.

The scale of the shift

This isn’t a hypothetical firms can wait out. In February 2026, Baker McKenzie cut between 600 and 1,000 business services roles globally. That’s the largest AI-attributed reduction the legal industry has seen to date. Clifford Chance made similar cuts the previous year, citing AI adoption directly, according to the Financial Times. Neither of those is a Canadian firm, but both operate here, recruit here, and set expectations that ripple through the Canadian market regardless.

Closer to home, a Thomson Reuters survey found that a large majority of Canadian legal professionals are adopting AI faster than they’re comfortable with — and that only a small fraction of firms are consistently disclosing that use to clients, a gap that has its own governance implications well beyond training.

None of this means junior lawyers are becoming redundant. A Law360 Pulse survey from March 2026 found both junior and senior lawyers believe AI could absorb responsibilities traditionally handled by junior associates — but American legal hiring data from the same period showed legal employment at a ten-year high, and law firms are still locking down top law students two and a half years before they start, in some cases paying five-figure sums simply to keep them off a competitor’s roster. That is not the behaviour of an industry that believes entry-level lawyers are becoming unnecessary. It’s the behaviour of an industry that hasn’t yet figured out how to rewrite their job description.

That’s the more accurate way to frame this. It isn’t a story about fewer junior lawyers. It’s a story about the same number of junior lawyers doing less of the work that used to teach them anything, unless firms deliberately intervene.

That instinct showed up again, from a different angle, when I interviewed Jacob Murad on a recent episode of my podcast, Lawyer Launcher: Behind the Bar. He’s a corporate-lawyer-turned-entrepreneur who now runs a 150-person professional services firm and mentors articling students through the Law Society of Ontario. Murad doesn’t think AI replaces junior lawyers; as he put it, someone still has to prompt it. But he’s already changing what his own juniors spend their time on, moving them off first drafts and research memos and into client intake calls and presentations instead, precisely because that’s the part AI still can’t do. His bet isn’t just that judgment needs deliberate reps. It’s that relationship skills do too. He’s spot on. AI can’t develop relationships and build trust.

What deliberate intervention actually looks like

Saunders’ answer isn’t to slow the technology down. It’s to rebuild the teaching around it. He argues firms need to double down on mentorship that happens in real time: bringing juniors into client meetings and boardrooms, not only handing judgment back through redlined documents by email. He also wants law schools brought into that redesign, so graduates arrive with some of this baked in rather than starting from zero.

I’d add a few practical ideas of my own, in the same spirit as the delegation checklist I laid out last column:

1. Treat AI fluency as a trained skill, not an assumed one — including how to interrogate an AI-drafted answer, not just how to prompt for it.
2. Delegate a few “slow” assignments per rotation, done without AI assistance, specifically because the friction is the point.
3. Have juniors review AI output against the source documents before a senior lawyer ever sees either version, so the verification is being practiced, not assumed.
4. Bring juniors into client conversations earlier and more often, even in a listening capacity, since that’s judgment-transfer that no drafting assignment ever fully replicated anyway. Let some of that time go toward intake calls, presentations, and other relationship-building work AI genuinely can’t do yet, rather than treating those as senior-lawyer territory.
5. Compensate and recognize the lawyers doing this coaching. Saunders is right that mentorship has long been treated as invisible labour. If it’s now doing more of the training work AI has hollowed out elsewhere, it can’t stay unpaid and unmeasured.

None of this requires firms to slow down their AI adoption, and none of it is really about AI at all. It’s actually about firms deciding, deliberately, what a junior lawyer’s first few years are supposed to build, and then reverse-engineering a plan, instead of discovering the gap five years from now when a mid-level associate can’t be trusted to run a file alone.

Recruitment gets the budget and the attention. Delegation quietly determines the return. And now the raw material that used to make delegation possible – the unglamorous, repetitive, judgment-building work – is the thing disappearing fastest.

Which leaves firms with a new question: if the reps are gone, what replaces them — and who, exactly, is responsible for helping students and juniors develop good judgement?