Rob Aberdein, an advocate of handing junior legal work to AI agents, has written in The Scotsman that every senior lawyer he knows learned the job through slow, repetitive work, and that this is the work agents absorb first. Reviewing documents, researching and drafting letters that come back covered in corrections were, in his words, never just cheap labour: “They were the curriculum.” He does not argue for slowing down. He argues that firms must design what osmosis used to deliver. That is the right question, and it lands on a profession whose training rules in England and Wales count the time juniors spend but leave its quality to the firm.
Strategic Insight: Aberdein calls it “the paradox of the agentic law firm”: “the agent performs the task, but the lawyer remains accountable”. If accountability needs competence, and competence came from repetition, then the efficiency case for agents depends on a training system nobody has yet built to replace it.
Why is the apprenticeship question arriving now?
Aberdein is general counsel at Hoxton Wealth, where he is also group director for legal and strategic growth, and runs Simpson & Marwick Group as managing director. He has previously argued that AI will reshape legal billing and the way legal work is done, and here he asks readers to assume he is right and that the change arrives quickly. His point is what follows from his own advocacy.
He is not alone in making it. Research from Positive Group, shared with City AM in June, suggested AI was dismantling law’s traditional apprenticeship model by automating document review and routine research, the tasks where juniors used to build their skills. It warned of “less requirement to go and interrogate sources”, leaving firms exposed to over-trusting what the tools produce. City AM also reported its finding that over 60 per cent of lawyers use AI day to day, for work including drafting and research. We covered that warning at the time.
The pipeline may also shrink before the training question is answered. Lord Neuberger, former President of the UK Supreme Court, was reported from Kuala Lumpur in October as saying that AI “could sharply reduce the recruitment of young lawyers” (our report).
| Figure or rule | Source | What it means for training |
|---|---|---|
| More than 60% of lawyers use AI in everyday tasks | Positive Group, via City AM | The tasks juniors learned on are already being shared with tools |
| Two years of qualifying work experience, full-time or equivalent | Solicitors Regulation Authority | The time requirement stands, whatever fills the two years |
| Exposure to a minimum of two competences | Solicitors Regulation Authority | A narrow role can still count as qualifying experience |
| Simulated legal services do not count | Solicitors Regulation Authority | Practice exercises on dummy matters cannot replace live work |
What does the work actually teach?
Aberdein’s argument is concrete. A junior with 500 leases behind them recognises an unusual one; a junior who has written one letter twenty times begins to see why the partner rewrites it each time. The knowledge is pattern recognition built by volume, and the volume is what agents remove.
His paradox follows from it. Someone reviewing a contract an AI produced needs a sense of what a sound one contains, where the pitfalls sit and when polished text hides a mistake. “But accountability requires competence”, he writes, and that instinct “doesn’t arrive with a practising certificate”. The risk he names is a generation who can run the tools without being able to tell when they are wrong. “That is not a technology problem. It’s a professional standards problem.”
Critical Context: The Positive Group warning describes the same failure from the other end. Its chief executive, Will Marien, said firms must equip staff “to overcome automation bias and over-trust”. On Aberdein’s argument, a supervisor who learned without the repetition would struggle to spot an agent’s confident error.
What does Aberdein propose instead?
He is candid that he does not have the full answer: “I don’t have it fully solved”. He rules out two responses. Slowing adoption is one, and inventing busywork is the other, because clients will not pay for it. His alternative is to “stop relying on osmosis and start designing how lawyers learn”. In practice that means three things:
- Earlier client exposure. Juniors joining client calls, drafting advice and seeing negotiations in real time.
- AI as sparring partner. Trainees pick apart an agent’s draft and hunt for its mistakes, instead of the agent replacing their attempt.
- Feedback as core work. Partners treat feedback as part of the job and keep a record of the work each junior has really seen.
“None of this is complicated. It is simply good supervision”, he concludes. In our view, one firm’s product comes close to the sparring-partner idea. Shoosmiths told City AM in June that Project Apollo, a contract-review tool built with Microsoft, would help junior lawyers “accelerate their learning”. Its chief executive, David Jackson, said the tool lets lawyers “not only see what amendments have been made, but most significantly, why”. The firm said every output is checked and approved by a senior lawyer (our report).
Reality Check: In our view, Aberdein’s measures are cheap to describe and expensive in partner hours. Earlier client exposure and structured feedback both draw on senior time, and he asks partners to treat feedback as core work, not an extra.
Who is affected?
| Group | How the shift reaches them |
|---|---|
| Trainees and paralegals | The tasks that built their judgment are the ones agents take on first |
| Supervising partners | Feedback and exposure tracking become work they are expected to do, not a by-product |
| Firm leadership | Short-term margin gains from agents sit against a thinner bench of future supervisors |
| Clients | Positive Group quotes one legal leader saying clients “use AI… but want everything human validated, cheaper, and with full liability” |
| Graduates choosing firms | Aberdein says more of them now ask what they will learn, and how fast |
What this means for training principals
In England and Wales the regulator sets the frame but not the content. The Solicitors Regulation Authority requires two years of qualifying work experience that involves providing real legal services, and states that “Simulated legal services also do not count – it must be real life experience.” On our reading, that matters for Aberdein’s sparring-partner idea: a trainee critiquing an agent’s draft on a live client matter is doing real legal work, but the same exercise on a training file would be simulation.
The SRA is also explicit that confirming a candidate’s experience “does not involve judging how well a candidate has been trained, the quality of their experience or the standard of their work”. Competence is tested through the SQE assessments instead. So the exposure tracking Aberdein recommends is not something the regulator does for firms. If it happens, the firm has to choose to do it.
Implementation Note: The SRA publishes a template for recording experience against its competences, while stating that it is not a regulatory requirement and is “a training and recording aid only”. Firms looking for a starting structure for Aberdein’s exposure tracking already have one.
Open challenges
Who pays for supervision. Aberdein says clients will not fund busywork, and the clients Positive Group describes want work that is human-validated and cheaper at once. That leaves open who funds the partner time his model needs. We looked at one firm’s proposals for funding development outside client work in September; the question remains open.
Who supervises the supervisors. Critiquing an agent’s draft only teaches if the supervisor can see what the trainee missed. On Aberdein’s logic, that depends on the judgment of partners who learned by the old route.
How many juniors there will be. Aberdein asks how firms can “take young lawyers with us” and train the next generation properly. Lord Neuberger’s warning is that AI could cut that hiring sharply. A smaller intake could make careful supervision easier to afford, or it could become the reason no one builds it.
Warning: ⚠️ A junior has already taken the blame when AI-assisted work went wrong in court. City AM reported that a High Court judge criticised Pinsent Masons at the end of May over letters produced with AI that a junior lawyer had sent to the court, and that contained false information (our report).
What to watch
- Graduate recruitment pitches. Aberdein predicts that firms able to show a structured, AI-literate approach to training will win better recruits. Watch whether firms start publishing what trainees will be exposed to, and how they will be supervised, in terms more specific than “you’ll pick it up as you go”.
- Trainee and graduate intake. If firms report smaller trainee cohorts in their next recruitment rounds, Lord Neuberger’s warning is landing first, and the training-design question becomes one for fewer people.
- Regulatory guidance on AI-supported work. Any statement from the SRA on how work done alongside AI agents counts towards qualifying experience would show whether the regulator intends to say anything about the content of training, not just its length.
- Who is named when things go wrong. Further court criticism where a junior is blamed for an AI error would suggest firms are adopting agents faster than they are building the supervision Aberdein describes.
Aberdein closes on a warning that the profession could build its most efficient firms ever “with nobody ready to lead them”. His article does not settle how to prevent that, and he says so. What it does is move the question from whether juniors will use AI to which skills lawyers must still have when AI handles everything else, and that is a question each firm will answer by what it chooses to fund.
Source: AI may mean death of legal apprenticeship by Rob Aberdein (The Scotsman, 2 November 2026). Training requirements from the Solicitors Regulation Authority’s qualifying work experience guidance. Positive Group findings as reported by City AM (June 2026); Shoosmiths details as reported by City AM (June 2026); Lord Neuberger’s remarks as reported by the New Straits Times (October 2026).
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. Read more analysis at Insights, or get in touch.