I’ve been thinking about which jobs are effectively finished because of advances in AI. In short, any work that is repetitive, has abundant public training data, and mostly maps language input to language output is at serious risk.

Law is the clearest example. That does not mean all lawyers are finished, but the impact is already visible: hiring of law graduates into junior roles has dropped sharply. The picture, however, is more nuanced.

Partner-level roles (including public-sector equivalents) will remain essential for as long as legal services are delivered to humans. These roles are fundamentally about sales, judgement, accountability and human interaction—none of which it is reasonable to delegate fully to AI without some implausible humanoid interface.

Even so, partners alone are not enough. Senior lawyers just below partner level do the real orchestration: validating work, making decisions, and ensuring quality. That layer is also essential.

The layers below that—junior and mid-level roles where people are not yet trusted to work independently and whose output requires senior review—are the most exposed. There is little doubt that AI will handle this work in the near future, faster and better than humans. The main reasons it has not yet happened are institutional caution and slow adoption of agent-based systems, both understandable in law. Once secure systems can guarantee confidentiality and a baseline level of quality, this work will move to AI.

It is an ideal use case for agents: contextualising a matter, retrieving minimal relevant statutes and case law, and coordinating specialised sub-agents under human supervision.

So what happens to junior and mid-level lawyers? They do not disappear entirely. Partners and seniors are human: they retire, move on, and occasionally die in post. Organisations therefore need a pipeline to replace them.

Assume a legal organisation loses around 10% of its partners and seniors each year. To replenish that talent, juniors must still be hired and developed, with attrition along the way. If roughly 10% drop out in each of the first three years, and 5% per year over the following seven years to reach senior level, then an organisation needs roughly double its annual senior attrition entering as juniors a decade earlier.

For example, an organisation with 100 senior lawyers might lose or promote 20 each year. To replace them, it would need around 40 juniors entering ten years prior. That implies roughly 300 junior and mid-level lawyers at any given time - this is a 500-person firm with 100 partners and 100 seniors included1.

This structure is far more top-heavy than in the past. Historically, firms needed at least three times as many junior and mid-level staff to grind through work before it was ready for approval. AI removes much of that need.

The implication is clear: we do not need to produce as many law graduates as we once did. Top law schools can continue at current levels, but many second- and third-tier providers are training graduates for roles that will no longer exist, forcing them to compete for a shrinking pool of positions against graduates from more prestigious organisations, yet incurring similar student debt and opportunity cost.

Is this really a disaster? Over a million law graduates are produced annually across the Western world. This was arguably a by-product of an increasingly complex legal system that required ever more human labour - until now.

This pattern will repeat across many industries. Organisations will still need to hire juniors, but primarily to develop future leaders rather than to supply large amounts of labour. Scale will adjust accordingly. The parallel with the industrial revolution is obvious: machines replaced most manual labour, not all of it. AI will do the same for knowledge work.

It’s a much bigger societal change than the move to computers from paper for knowledge work, much bigger than the effect of the internet on work. This is the reason why it may not be a bubble: some investors are counting brass tacks at specific companies… others are thinking bigger, towards the multiplier effect of a knowledge revolution on the economy as a whole and what that’s worth.