The pyramid is crumbling. Where do your juniors learn now?

AI is automating the work that taught junior professionals how to think. Firms that get this wrong may not notice until it’s too late.

A firm can produce answers faster than ever. But what happens when it has too few people who know whether those answers are right?

This technology is moving so quickly it’s creating a succession problem, and it is arriving faster than most partners think.

The unspoken bargain that built the profession

For generations, junior professionals wrote the first draft, prepared the workpaper, did the precedent search. Seniors reviewed, challenged and corrected. The client received the finished product, and in the process of producing it, the junior learned the fundamentals of their work.

The pyramid was never just a pricing model. It was the training programme. The pyramid did more than underpin the economics of professional services. It was also how generations of professionals learned their craft. And AI is starting to hollow out its bottom layer.

Tasks that once took hours now take minutes. The processes for first drafts, document review, technical research and analysis are all being reshaped. For firms under margin pressure, the efficiency case is obvious.

But those tasks were dual-purpose. They taught people new to the profession how to think.

Early evidence suggests AI adoption is changing the balance between junior and senior roles, even as firms continue to hire. Research from King's College London and Stanford University, covering 41 countries, found that firms adopting AI increased overall recruitment by around 3.3 percent. But the gains were uneven: senior employment rose by 6.7 percent, while junior employment fell by 2.5 percent. The study is not specific to professional services, and AI is not necessarily the only factor behind these changes. What the findings suggest is a workforce increasingly weighted towards experienced professionals, raising an important question about how firms will develop the next generation of expertise.

The real concern goes beyond junior headcount. If AI takes on the work that once trained junior professionals, how will they develop the expertise expected of tomorrow's partners?

One question, two very different outcomes

A junior tax adviser is assessing a complex client position.

Traditionally, they would read the legislation, work through the guidance, examine previous opinions, weigh the alternatives and defend a conclusion to a senior colleague. Slow.

Sometimes frustrating. But it built the ability to apply knowledge to circumstances nobody had seen before.

Now imagine AI produces a persuasive answer in seconds, citing earlier engagements and internal technical papers. The junior has an answer. But do they know if it is right? Whether it is, or isn’t, do they know why?

Here is where the risk gets sharper than a skills gap.

What if the opinion the AI relied on was superseded last year? What if it came from another client engagement, and carried confidential reasoning the junior was never authorised to see? What if three versions of the same advice sit in three locations, and the system retrieved the weakest one because it was the easiest to find?

The junior may not spot it. The answer could read confidently either way. And the partner signing it off may be relying on a foundation nobody has adequately checked.

Professional judgement has always depended on knowing the limits of what you have been told. Remove the experience that builds it, then feed people unreliable material at machine speed, and you do not necessarily get faster expertise. You risk having mistakes delivered at greater speed and with more confidence.

The profession already senses this. The Thomson Reuters Institute's Future of Professionals Report 2026 found that 48 percent of professionals surveyed were concerned about AI's negative impact on the development of independent professional judgement, and 71 percent believed early-career professionals needed structured support from experienced colleagues to build the skills that AI risks displacing.

Apprenticeship doesn’t end. It must be redesigned.

If juniors spend less time producing the first draft, they need more chances to interrogate previous decisions, challenge assumptions and understand the reasoning behind the firm's best work.

That makes institutional knowledge an essential part of the new training ground: previous engagements, technical opinions, regulatory interpretations, advice that was challenged, defended and refined. It is also the information AI systems connected to the firm's knowledge repositories can draw on to produce answers grounded in its own expertise.

The quality of your knowledge shapes two things at once: how reliable your AI is, and how well your next generation learns. A messy information estate does not just risk producing poor answers. It can reinforce poor judgement.

Firms hold decades of valuable knowledge, but how much of it was organised with learning or AI-powered reuse in mind? In all likelihood, it sits across disconnected repositories, in multiple versions, stripped of the context that made it valuable. A document records the final recommendation but not the alternatives considered, the assumptions behind it or why a senior professional rejected a different approach.

Handing that estate to AI does not fix the problem; it industrialises it. AI won’t fix weak knowledge, it will scale it. Information that is outdated, duplicated, or lacks context gets retrieved and reused everywhere, spreading the same errors across multiple engagements.

A knowledge repository cannot replace a mentor. But it can give the mentor and the junior a far better foundation to work from. And that matters, because a graduate who can retrieve yesterday's advice faster has not necessarily learned how to give tomorrow's advice.

What good looks like

This is not a call to slow down. It is a call to build the foundation the next generation of professional development depends on. Three moves matter.

1. Make context and lineage visible on everything.

A junior should see what the engagement was, when the advice was given, which version was approved and what assumptions shaped the conclusion. Where it matters, the review history and the reasoning behind the decision should be preserved alongside the document.

Provenance helps establish whether information can be trusted. Reasoning helps professionals decide whether it should be applied again. Both matter.

2. Enforce access before the AI reads, not after.

Need-to-know rules, ethical walls, client confidentiality and cross-border restrictions exist for good reasons. An AI system that retrieves information across an ethical wall can expose material to a user that isn’t authorised to see it.

Security must be enforced against the underlying content and respected by AI-powered retrieval, so every answer, whether it serves a client, supports a decision, or teaches a junior, draws only on information that person or agent is authorised to access.

3. Add the human layer on top.

Give juniors AI-generated work to challenge before they see how a senior approached the same problem. Let them study real engagements for the reasoning, not just the output. Make experienced professionals accountable for teaching how to recognise a convincing answer that is nonetheless wrong.

None of this works if talent, knowledge, technology and risk operate in isolation. Developing the next generation of professional expertise requires shared accountability across all four.

The question for your next leadership meeting

The pyramid will keep evolving, and that is not necessarily a bad thing. AI can free people from repetitive work and move them to higher-value thinking sooner. But only if firms take deliberate steps to replace the learning opportunities that automation removes.

That means treating institutional knowledge as more than a repository of documents or even as a source of information for AI. It is part of how a firm preserves its expertise, transfers its experience, and enables its future leaders to develop the professional judgement on which their continued prosperity depends.

The firms that succeed will not be those that preserve the old pyramid or dismantle it fastest. They will be those that find new ways to develop expertise while making the knowledge beneath it trustworthy, governed and understood in context.

Ask your leadership team:

If AI wrote our first drafts tomorrow, what would our juniors learn from, and could we prove it was safe to learn from?

If the room goes quiet, you have found the gap.