After decades of building elaborate credentialling systems, structured curricula and high-fidelity simulation suites, the UK government committed £725 million earlier this year to move people faster into supervised, work-embedded learning – cutting apprenticeship approval times from 18 months to as little as three months. The stated ambition, published in the GOV.UK release, is to get two-thirds of young people into higher-level learning or apprenticeships. It’s a substantial investment in the oldest model of professional formation that exists – which is either evidence that the model works, or a quiet tribute to how little the decades of alternatives actually moved the needle.
In surgery, commercial aviation, civil engineering and law, that shift makes explicit what practitioners already know: formal education produces qualified entrants, but expert practitioners emerge through repeated exposure to live, variable, high-pressure work. Credentialling systems can define entry standards and test procedures; what they structurally cannot encode is the contextual judgement that forms when reality diverges from the plan – and no redesign of curricula or simulation fidelity changes that.
What the Classroom Cannot Encode
Expert knowledge operates in two registers. The first is teachable, testable, and increasingly simulable: procedure under known conditions, standard scenarios, defined responses. The second forms through practice, resists direct instruction, and transfers poorly from a lecture slide. That’s not a failure of curriculum design; it’s a structural feature of expert work.
No curriculum can reproduce the full combination of variables a practitioner encounters in live work. Techniques can be documented; the reasoning that governs technique selection when conditions are unusual is context-dependent and hard to encode. Civil engineering makes this concrete. Structural analysis, materials science and load calculations are fully teachable, yet the risk intuition a graduate develops on a live infrastructure project forms beside a chartered engineer making real decisions. Northumbria University’s civil engineering degree apprenticeship follows that logic, deliberately blending university study with hands-on infrastructure work within a single pathway.
The profession’s own competency standard makes the acknowledgement formal. That professional judgement develops in supervised practice – not additional classroom hours – is precisely what the Engineering Council, the UK regulator for professional engineering registration, states in UK-SPEC: “It is how potential Chartered Engineers learn to apply their knowledge and understanding and begin to apply professional judgement.” A standard-setter confirming that its standards cannot, by themselves, complete the formation they define is worth pausing on.
The structures that reliably transmit expert capability share a common logic – enough real cases, close enough to a practitioner making live decisions, with a structured obligation to reflect on why one path was chosen over another. What formal systems cannot resolve is how much of that exposure is sufficient, or what happens to the formation of judgement when the caseload or the supervision thins.
Commercial Aviation – Reliability as Measurable Evidence
Commercial aviation has pushed simulation further than almost any other profession. Full-motion simulators and procedural trainers rehearse engine failures, low-visibility approaches and defined emergencies with high fidelity, and recurrent checks have genuinely lifted procedural safety. What they cannot reproduce is the situational judgement expert captains exercise when weather, tight turnarounds and network knock-on effects collide – the kind first officers absorb by flying beside them on live line operations.
The U.S. National Transportation Safety Board’s (NTSB) report on Colgan Air Flight 3407 documents what the absence of that judgement looks like: the captain made an inappropriate response to a stick-shaker stall warning and the aircraft entered an unrecovered stall, with contributing factors including failures in airspeed monitoring and non-adherence to sterile-cockpit discipline. That’s the failure mode; Virgin Australia’s operating record offers the other reading.
BITRE data record an 86.9% on-time departure rate and a 99.2% completion rate for Virgin Australia in May 2025. The airline planned more than 19,000 flights and around 3.2 million seats over the 2025 Christmas school holiday period, meaning individual cockpit and operations decisions aggregate into network-wide reliability that shows up in the numbers. Dave Emerson, Chief Executive Officer of Virgin Australia Airlines Pty Ltd, and his operations leaders attribute this “trifecta” of on-time departures, on-time arrivals and low cancellations to coordinated judgement by operations controllers, cockpit and cabin crews, and ground teams on the day of operation – built on, but not replaced by, simulator training, manuals and checklists.
The U.S. Federal Aviation Administration’s powered-lift Special Federal Aviation Regulation creates a training and certification framework for electric vertical take-off and landing (eVTOL) aircraft that allows more training in high-fidelity simulators while still requiring most pilot training in real aircraft. The regulator’s own rulemaking language makes the reasoning plain, referencing live operations in the National Airspace System (NAS) and citing “exposure and experience of operating an aircraft in the NAS, all of which cannot be fully replicated in a simulator.” Even where simulation is expanded, the framework preserves live operations as the irreducible site of competence formation.
Law – The Institutionalised Acknowledgement
In common-law systems, stages after law school – pupillage, training contracts, clerkships – are not leftovers from a pre-digital era. They exist because attempts to replace supervised practice with more classroom time keep meeting the same reality: legal knowledge and legal judgement differ, and applying doctrine to ambiguous disputes with real clients requires guided exposure to live matters.
The Supreme Court of California made that logic structural, amending its rules for the Law Office Study and Certified Law Students programmes so that some participants can join both, including those studying under a judge – provided supervisors manage conflicts of interest and the student does not appear before that same judge when representing a client. The result is more supervised practice routes, not fewer.
Axios reported that artificial intelligence is eliminating some of the entry-level tasks through which junior associates develop their first instincts about argumentation, risk and client reality. The timing has a certain symmetry: courts formalising more supervised practice pathways at the same moment AI is removing the work those pathways depend on to be formative. The Carnegie Foundation’s report Educating Lawyers: Preparation for the Profession of Law frames legal formation as an apprenticeship process, distinguishing doctrinal classroom knowledge from the practice-based reasoning and professional identity that only develop through real matters. Together, that analysis explains why eroding entry-level work isn’t a neutral efficiency trade – but it doesn’t answer what architecture delivers the formation most reliably.
Surgery – Volume, Variability, and the Fellowship Model
A dose–response meta-analysis in spine surgery found that higher surgeon volume is associated with lower postoperative morbidity and mortality, along with shorter length of stay and fewer readmissions. Read as a systems signal, it suggests that when real-case throughput is low, the cost is not only slower skill formation but potentially higher patient risk.
Dr Timothy Steel’s Spine Surgery Fellowship, run over six to twelve months in collaboration with St Vincent’s Private Hospital and Concord Hospital, is built around that volume requirement. Fellows receive supervised exposure across minimally invasive decompression, open and percutaneous fusion, disc replacement and vertebral reconstruction, assisting in approximately 500 procedures each year under his supervision. That density matters: the meta-analysis reports that high volume correlates with better outcomes, reinforcing the idea that exposure itself is doing formative work – which is precisely what the fellowship is structured to deliver.
The fellowship’s teaching resource is the practice itself. Steel is a neurosurgeon and minimally invasive spine surgeon whose consultant appointment dates from 1998; the more than 8,000 minimally invasive procedures and over 2,000 complex spine operations recorded on his specialist listing are what give fellows proximity to decision-making at the field’s edge, not a separate qualification alongside it. Fellows are also required to complete two research projects to final-draft level – converting operative decisions into written, examined reasoning rather than letting experience settle as unexamined habit.
That architecture – volume, variability, supervised exposure and structured reflection – not the simple passage of six to twelve months, is what determines whether a fellow leaves with the contextual intraoperative judgement needed for independent practice when cases diverge from standard patterns.
The Substitution Question
Framing the impact of technology on professional training as a choice between simulation and apprenticeship misses the real issue. Procedure under known conditions is increasingly well handled by simulators, and the powered-lift training framework shows that simulation can safely absorb more training hours where operating envelopes are tightly defined – even as real-aircraft time remains mandatory.
What the evidence across these fields does not support as substitutable is the live, consequential work through which judgement forms. The AI-driven erosion of junior-lawyer tasks, the volume–outcome relationship in spine surgery, the reliability gap between procedural proficiency and live operational judgement in aviation – these are different fields reading the same underlying pattern. Expert capability tracks supervised exposure to real variability, not the sophistication of the inputs alone.
The decision to fast-track apprenticeship approvals and fund expanded work-based training reflects that recognition at policy level. But funding the sprint back to the oldest model is only useful if the structures receiving those resources are actually built to transmit judgement – not just to log supervised hours. In high-consequence fields, the distinction between the two is the difference between a practitioner who can follow a protocol and one who knows what to do when the protocol runs out.