How remote work and AI are quietly dismantling the way professionals are made.
Apprenticeship was always informal
The trades figured apprenticeship out a century ago with indentured years, structured progression, and formal certification. There is no shortcut, and sparkies, plumbers and chippies emerge as competent tradespeople because the system forces the time and proximity that produce one.
Professional services never built anything comparable. There is no formal apprenticeship for a graduate lawyer, junior analyst or first-year consultant, only an implicit version where a few years of routine work like drafting, reviewing, modelling and summarising, done under the eye of someone senior, slowly built the pattern recognition that made them useful. It worked, but only while two conditions held: routine work for juniors to do, and seniors close enough to see them do it. Both are now in question.
Remote work removed the watching
Apprenticeship in professional services was never really about formal training, it was about proximity. Overhearing how a senior handled a difficult client call, watching them mark up a draft in real time, being pulled into a meeting because you happened to be at your desk. Judgement transferred in the margins, through corridor conversations, coffee runs, and the senior leaning over your shoulder to point at something on screen. Nobody had to design it, because being in the same place eight hours a day did the work for them.
Remote work removes nearly all of that, and mentorship now has to be deliberate, scheduled and explicit. The Big 4 will point to their graduate programmes with rotations, structured training and performance frameworks, and they are not wrong, but those programmes teach methodology and technical skills. The judgement bit was always informal, always carried by proximity, and that is what broke. What happens instead is mechanical: juniors get tasks via Teams, deliver them via Teams, get feedback via Teams. The work product gets reviewed but the thinking behind it does not transfer, and the pattern recognition that compounds into judgement stays with the senior.
AI is taking the doing
What AI does best is what juniors used to do: first-pass drafting, document review, routine analysis, decks from templates, boilerplate code, research summaries, meeting notes. That work filled the first two or three years of a professional career, and it was not just output, it was development. Judgement comes from doing it, getting corrected, doing it again, and slowly building the instinct that makes someone senior. Pull that work out from under the next generation and juniors stay juniors, never becoming seniors.
The signal is already showing in the labour market. A 2025 Stanford study found that early-career workers in the most AI-exposed occupations have seen a roughly 13% relative decline in employment since late 2022, even as their older colleagues in the same jobs held steady. Companies are not firing so much as not hiring or replacing, which is quieter than a layoff wave but more structural. Headline unemployment stays manageable because incumbents keep their seats, while the bottom of the ladder is quietly being removed.
Two forces compounding
Either force alone would be a challenge, but together they are a structural problem. Even where juniors are hired, they are not getting the experience the role used to carry. The work product looks fine, but the slow accumulation of judgement that made someone senior in five or ten years is not happening. Today's seniors were trained under the old model and are still capable, still in the room. The problem shows up in a decade when there is no one ready to replace them, by which point the cause is too far back to attribute and the fix is too far ahead to deliver in time.
What this means for enterprises
When enterprises replace junior work with AI, they think they are making a productivity decision when in fact they are making a succession decision. The savings show up this year, the missing seniors show up in a decade.
Remote work is not the villain here, and it is not going back either. Employees value the autonomy, and most are not willing to return to nine to five in an office. That is fair, but the incidental mentorship that used to happen by sharing space is not coming back with them, and pretending otherwise is the problem. Mentorship has to be designed now, deliberately:
- Pairing juniors and seniors on real work.
- Scheduled time where the senior narrates their thinking rather than just reviewing output.
- In-person days used for the work that actually benefits from being together.
- Seniors measured on developing people rather than only on delivering work.
None of it is exotic, just no longer optional. Juniors also still need to do the routine work, even when AI could do it faster, because the doing is how they develop.
What AI can give back
AI is breaking apprenticeship by taking the doing, but AI built by the people who would have been mentors could carry back some of what proximity used to deliver. A senior's accumulated judgement, how they review a draft, what they look for in a contract, the heuristics they have built over twenty years, can be captured and made available to juniors as an agent that scaffolds their thinking rather than replacing the senior. The agent can critique a draft the way the senior would, explain the reasoning behind a heuristic the senior might not even articulate consciously anymore, and be patient in a way a senior under load cannot always be.
It is not a full substitute. A lot of what makes a senior a senior is tacit, and an agent trained on what they can say will miss what they cannot. Agents do not get tense in a difficult conversation, do not read a room, and do not have skin in the game. But it is not nothing either. It carries part of the load, scales the senior's attention, and only works if seniors invest the time to build and curate it, which is itself a form of mentorship being designed back in. This is the more honest version of augmentation: not AI replacing the work that built the next generation, but AI built by one generation to help develop the next.
The problem worth solving
The headline unemployment number will stay manageable for years while the system that produced professionals is being quietly dismantled underneath. Remote work ended learning by exposure and AI is ending learning by doing, neither decision made with apprenticeship in mind, both reshaping it anyway.
The organisations that handle this well will recognise that apprenticeship was always a system, that the system has been damaged, and that rebuilding it requires intention rather than just hiring graduates and hoping. The ones that do not will discover, somewhere around 2035, that they have run out of seniors and do not know why.
Responsible AI is not just about model risk, privacy and bias. It is about whether the organisations and governments deploying AI take responsibility for what it does to the people who work, and for the professionals we should still be developing. Sustainable use of AI means sustaining human capability alongside it, not in spite of it, and organisations and governments both need to be accountable for both.
Further reading: Enrique Ide, Automation, AI, and the Intergenerational Transmission of Knowledge (2025). An economic model showing that automating entry-level tasks can slow the transfer of tacit expertise from experts to novices, and reduce long-run growth, even when entry-level jobs themselves are not cut.
Shaun Crouch is the founder of Pragmatix, a Brisbane-based digital transformation consultancy. His career spans programming, solution architecture, and enterprise architecture across healthcare, government, and financial services. Pragmatix helps organisations adopt AI responsibly, sustaining human capability alongside the technology, not in spite of it.
Pragmatic Agentic AI, Shaun's free eBook on building autonomous AI systems pragmatically, is available at pragmatix.com.au/agentic-ai. The Pragmatix Advisory Portal at portal.pragmatix.com.au provides access to expert AI advisors for governance, cyber security, and solution architecture.