The expertise debt no one is counting

The firm announced it was cutting its 2025–26 graduate intake from 104 positions to 72 – nearly a third – and reached for none of the usual exits. No “market recalibration”. No “strategic workforce alignment”. Chief people officer Rachel Banks told The Australian Financial Review plainly: AI is improving efficiency in the routine work graduates traditionally start on, and that has led the firm to take a more targeted approach to hiring.

That candour is overdue. But the profession’s response – a mix of concern for graduates and quiet approval of the efficiency logic – is reading the wrong signal entirely.

This is not a workforce story. It is an expertise story. And the implications run deeper than most managing partners are currently modelling.

The surface reading is wrong

The prevailing interpretation is this: AI has automated repetitive, low-value work that junior lawyers once performed. Firms need fewer juniors. Demand for senior, complex legal advice remains strong. Fewer graduates, better deployed. The logic is clean.

The conclusion is premature.

VIEW ALL

The graduates who do secure roles will be deployed faster into sophisticated matters. Firms frame this as an opportunity. In practice, it may be the acceleration of inexperienced practitioners towards complexity they are not yet equipped to navigate.

First-order implication: The graduate market has permanently contracted

The most visible consequence is already dominating the conversation. Fewer positions. More competition. Greater uncertainty for law students who have structured years of study around a career pathway that has quietly narrowed.

For the profession, this means the intake funnel that has historically produced a broad range of future lawyers – across commercial practice, public interest work, government, in-house, and the bar – is contracting at the source. MinterEllison has broken the silence. The economics that drove their decision apply to every firm running the same technology stack. The question is not whether others will follow. It is when, and how honest they will be about saying so.

First-order: Structural contraction in early legal careers. Significant. But the least dangerous of the three consequences.

“Document review and repetitive drafting were never just billing opportunities. They were the crucible in which junior lawyers developed professional instinct.”

Second-order implication: The expertise pipeline is being hollowed out in silence

The more serious problem will not surface for five to seven years. By which time, fixing it quickly will not be possible.

Senior legal expertise is not a credential. It is accumulated capacity – built through thousands of hours of work that felt routine at the time, through noticing anomalies that supervisors had to point out before you learnt to find them yourself, through getting things slightly wrong in low-stakes matters, and developing the judgement to distinguish a genuinely complex situation from one that merely looks complex. That formation process cannot be compressed. It cannot be simulated. And it cannot be skipped.

Harvey AI has recently expanded its law school program to Australia, partnering with the University of Sydney Law School and UTS Faculty of Law. Tomorrow’s graduates will be fluent in AI legal tools from their first day of practice. That is good. The risk is not their familiarity with the tools. The risk is that the tools begin performing the cognitive work that used to form them.

What the current model assumes – that graduates can be productively fast-tracked into sophisticated mandates – confuses deployment with development. A junior lawyer assigned to complex work earlier is not the same as a junior lawyer who has developed the capability to handle complex work. The gap between those two things is where professional risk lives. Quietly. For years before it becomes visible.

Human-led AI, not AI-led humans, is the goal every firm says it is pursuing. But human-led AI requires humans who have formed sufficient expertise to lead. If the foundational years that build that expertise are systematically removed from the development pathway, the leadership capacity that firms are counting on will not exist when it is needed.

Second-order: A senior talent crisis developing silently in the mid-2030s, entirely absent from current workforce modelling.

Third-order implication: The profession’s duty of care is at risk

The deepest implication is the least comfortable to name, because it requires the profession to ask a question about its own obligations.

Professional conduct obligations under the Legal Profession Uniform Law require practitioners to bring genuine competence to the matters they handle. Competence is not binary. It is developed through exposure, tested through practice, and maintained through deliberate repetition. It is experiential. It cannot be conferred by a tool, however sophisticated.

If the next generation of senior practitioners is formed primarily through AI supervision rather than foundational practice, the profession faces a structural question it has not yet asked publicly. Who checks the AI output when the supervising practitioner has never performed the underlying task manually? Who recognises the edge case when their entire professional formation has been shaped by tools optimised for standard patterns? And who carries the liability when the judgement call – the one that ultimately belongs to a human – turns out to be wrong?

The third-order implication is a distributed erosion of professional judgement across the Australian legal profession. Invisible at the level of any individual firm. Systemic at the level of the profession. Irreversible quickly once it has taken hold at scale.

Problems first, platforms second. The platform has changed. The professional obligation has not.

Third-order: Distributed erosion of professional judgement across the profession – invisible at firm level, systemic at profession level.

“The firms that navigate this well will not be those that adopt AI fastest. They will be those that build the human conditions for expertise alongside the technological ones for efficiency.”

The question every firm needs to answer

At GPS-AI, we work with professional services firms navigating exactly this tension. The question we hear most often is no longer “should we adopt AI?” – that decision has been made at every serious firm. The question now is: “What do we protect?”

The firms approaching this well are not slowing down their AI adoption. They are doing something more demanding: designing an explicit expertise formation strategy alongside their AI deployment strategy. They are mapping the cognitive development work that used to happen organically through routine tasks – and asking where that development now occurs. They are building deliberate structures – supervised review, mentored reflection, structured exposure to complex reasoning – to replace what the AI has absorbed.

GPS-AI operates on a core principle: AI is 20 per cent technology, 80 per cent psychology. The firms that handle the transition well will be those that understand the psychological and developmental dimensions of expertise formation – not just the efficiency gains that come from automating routine work.

3 questions managing partners need to answer now

MinterEllison’s announcement is a signal. The firms reading it as a prompt to consider reducing their own intake numbers are asking the wrong question. The firms reading it as an invitation to examine what expertise formation actually requires in an AI-augmented practice are asking the right one.

Three questions that warrant genuine answers – not policy documents:

Question 1

If AI is performing the foundational work that used to train your junior lawyers, what is your explicit strategy for developing the professional judgement that work used to produce?

Not your technology strategy. Not your efficiency metrics. Your expertise formation strategy – the deliberate, structured pathway from graduate to trusted adviser in an environment where the traditional formation work no longer happens at the same volume or pace.

Question 2

In seven years, when your current senior cohort begins to retire in material numbers, where is the next generation of deep expertise coming from – and can you trace the specific development pathway that produces it?

Workforce planning that does not extend beyond a three-year horizon is not planning. It is scheduling.

Question 3

What is your exposure if a practitioner formed primarily through AI-supervised practice makes a judgement error on a complex matter – and the investigation leads back to your firm’s development model?

This is not a hypothetical risk. It is a liability question. Better examined now, in a boardroom, than later in a courtroom.

The expertise debt accumulates quietly

Justice Michael Kirby observed that the internet cannot teach wisdom and judgement. Neither can AI. That observation is not a reason to slow the adoption of AI in legal practice. It is a reason to be deliberate about what sits alongside it.

AI will not replace lawyers. The MinterEllison announcement confirms this – demand for complex legal work is growing, not shrinking. But a profession that automates the conditions under which expertise is formed, without replacing them with something equally rigorous, is not becoming more capable. It is becoming more efficient at the same time as it accumulates an expertise debt that will come due on someone’s watch.

The question for managing partners is not whether AI changes the economics of graduate hiring. It plainly does. The question is whether the profession that emerges from those economics will still know how to do what clients need it to do – and whether the humans at its centre will still be genuinely leading it.

And it needs answering before the silence becomes permanent.

Andrew Cooke is the founder and managing director of GPS-AI (Growth & Profit Solutions AI).