If you take one thing from this article, take these four moves:

• Use AI for the routine work, but keep your people learning the judgement behind it.
• Move your experts toward the work AI cannot do: judgement, relationships, oversight.
• Reskill deliberately. Build AI fluency and supervision skills across the firm.
• If roles change, meet your Fair Work consultation obligations before deciding.

Everything below is optional depth. Read on for the evidence, the Australian picture, and the practical detail, including where the law applies.

Digging Deeper

Below sits the supporting case. It covers what is changing, the pipeline problem at the centre of it, and how to build a human-AI workforce that grows expertise rather than eroding it.

What is actually changing for professional firms

AI is now genuinely capable at routine cognitive work. It drafts, researches, summarises, reviews, and reconciles, quickly and at low cost.

The scale is real. McKinsey estimates that more than half of current work hours could in theory be automated. The point is not mass job loss. It is that roles shift toward judgement, validating, prompting, and managing the work that AI produces.

Adoption has moved fast. McKinsey found that employees reporting AI use at work rose from about 30 per cent in 2023 to 76 per cent in 2025. Humans and AI are already working side by side in most firms. A human-AI workforce is no longer a forecast. It is the present.

For an accounting, law, or consulting firm, here is the catch. The most automatable work sits at the junior end. That is where the human-AI workforce question bites hardest.

Picture the typical junior brief. Pull the documents, summarise them, draft the standard advice, reconcile the accounts, check the precedents. Much of that can now be produced in minutes. A human-AI workforce changes the junior’s day first.

The pipeline problem no one is pricing in

Here is the issue that rarely makes the business case. Juniors learn judgement by doing the routine work. This is the pipeline problem at the heart of every human-AI workforce decision.

Reconciliations teach how the numbers fit together. Document review teaches what matters and what does not. Research teaches how the law actually moves. The grunt work is also the apprenticeship.

If AI does that work, the apprenticeship can quietly disappear. McKinsey puts it well with a simple analogy. A calculator does not remove the value of learning maths. Some foundational ability comes only from doing the thing yourself.

Think of a first-year accountant. A few years ago, months of reconciliations and basic returns slowly taught them how a business actually works. If AI completes that work, the lesson does not happen on its own. A human-AI workforce has to design it back in.

The Australian data shows the early shape, and it is mixed. Graduate job postings fell almost 15 per cent in 2025, and sit about 35 per cent below their 2023 peak, though they began stabilising in early 2026. The evidence that AI is the main cause is far from settled.

Jobs and Skills Australia found that entry-level roles are more likely to transform than disappear, but flagged agentic AI, which handles structured routine tasks, as one to watch. Some Australian firms already hire fewer juniors and more senior staff.

The honest risk is long term. As one Australian analysis put it, without entry-level roles, where do tomorrow’s experienced professionals come from? For a profession that already faces skills shortages, thinning the base now stores up a senior shortage later.

The legal sector shows the nuance. Some large Australian firms keep recruiting graduates while investing heavily in AI. Yet junior lawyers worry that research, long a training ground, is the most exposed. Both things can be true at once.

Redesign how people learn, do not just cut the work

The answer is not to ban AI from junior tasks. It is to rebuild learning on purpose, so a human-AI workforce still produces real expertise.

This is very doable. The shift is from juniors producing the work to juniors directing and checking it, which can teach judgement faster, not slower.

Practical moves for a firm:

• Have juniors critique and correct AI output, not just receive it.
• Walk them through the reasoning, not only the finished task.
• Pair AI use with real mentoring, review, and feedback.
• Make supervising AI an explicit, taught skill.

McKinsey describes the new core skill as working effectively with machines, validating and prompting their output. In a human-AI workforce, that skill carries a genuine premium. Teach it early.

A simple example shows the shift. Instead of asking a junior to draft a memo from scratch, ask AI for the draft, then have the junior find its three weakest points and fix them. They learn judgement by interrogating the work, which is the skill that lasts.