The Future Doctor: Part Clinician, Part Editor, Part Systems Thinker

Medical training still quietly assumes a particular professional: a person who carries knowledge in their head, produces clinical text by hand, and works inside systems someone else designed. Examinations reward retention. Rotas reward throughput. The system's design — the workflows, the software, the failure modes — is treated as weather: something doctors work in, not on.

Each of those assumptions is now decaying at a different speed, and the doctor being assembled by the next decade looks like a hybrid that training hasn't caught up with: still a clinician at the core, but increasingly also an editor — a professional reviewer of machine-produced clinical material — and a systems thinker, because the systems have started making clinical decisions and someone clinical has to be looking at them. This isn't futurology. All three shifts are observable now. The only question is whether the profession develops the second and third skill sets deliberately or by accident.

The editor

Start with the most concrete shift. As machine-drafted material spreads through clinical work — notes, summaries, letters, suggested problem lists — the clinician's daily relationship to text inverts. Less authorship; more review. And review, as every editor and every consultant counter-signing a junior's letter knows, is a genuine skill with its own failure modes, not a diluted form of writing.

The editor's core competence is knowing where the errors live. Machine-drafted clinical text fails differently from human-drafted text: fluent fabricated specifics, smoothed-over distinctions, and — most dangerously — plausible omission, the thing the draft never said and doesn't know it never said. Catching what's absent is a different cognitive act from checking what's present; it requires holding your own model of what the document should contain and auditing against it, rather than reading along nodding. Reading along nodding is precisely what fluent drafts induce.

The editor also owns the output. The signature under a machine draft is a clinical act — the moment the words become the clinician's words, with everything that implies at review, at handover, in court. A profession of editors needs explicit norms about what 'reviewed' means: read fully? checked against the encounter? verified the numbers? Today this is left to individual conscience under time pressure, which is to say it is being decided by the rota. That's an answer too, just not a chosen one. ('Edited under pressure' is how the EHR era's worst habits got installed; the profession has one chance to set better defaults this time.)

What training would look like if this were taken seriously: deliberate practice at adversarial reading — drafts seeded with the actual machine failure modes — the way other safety-critical professions train their reviewers. Nothing about it is mysterious. It just isn't taught.

The systems thinker

The second shift is larger and less visible. Clinical software used to store and move information; increasingly it shapes decisions — what gets flagged, ordered, suggested, prioritised. Once a system participates in clinical reasoning, its design is clinical practice. The alert threshold is a clinical judgment, made by whoever set it. The triage logic, the defaults in the order set, what the dashboard makes visible and what it quietly buries — all clinical judgments, frozen into configuration, applied to every patient who passes through.

If clinicians can't reason about systems, those judgments get made without clinical reasoning. That's not a hypothetical — it's a fair history of two decades of health IT, and a preview of the AI deployment decisions being made in procurement meetings right now.

The systems thinker's repertoire isn't engineering. It's a way of interrogating any process or tool: Where does this fail, and what happens then — does it fail loudly into a fallback or silently into a gap? What behaviour does it actually incentivise under load, as opposed to intend? Where does the workload it 'saves' actually go? What does it do to the patient it wasn't designed for? These are recognisably clinical instincts — anticipating deterioration, asking what's being missed — pointed at the system instead of the patient. Emergency physicians, who spend their working lives watching systems fail in real time and improvising around the failures, come by this honestly; the profession at large needs it formalised: in training, in job plans, and in who sits at the table when deployment decisions get made.

The clinician — what stays load-bearing

Naming what grows is only honest alongside naming what holds. Three things, none of them sentimental.

Presence with the patient: eliciting the history that wasn't volunteered, examining a body, reading the person as well as the complaint — the input layer machines either can't access or can't be trusted with. Judgment under irreducible uncertainty: the weighting of risk for this patient, with these values, when the data has run out and the decision still has to be made tonight. And responsibility: being the named human who stands behind the call. If anything, the editor and systems-thinker roles make this third element more central — in a clinical environment increasingly full of suggestions, drafts, and flags, the scarcest commodity becomes someone accountable who decides.

The point of the hybrid framing isn't that doctoring shrinks. It's that these core capacities now need defending deliberately — because the editor's time pressure and the system's automation both erode them quietly. The skills that stay load-bearing are exactly the ones that atrophy without deliberate use.

The uncomfortable bit for training

Map the three parts against what medical formation currently optimises and the mismatch is stark. Years of curriculum on knowledge retention — the component AI has most thoroughly commoditised. Almost nothing on adversarial review of clinical text. Systems thinking confined to an audit project and a quality-improvement module taught as bureaucracy rather than as the safety discipline it is. Communication and uncertainty — the load-bearing clinical core — still examined as a checklist performance rather than developed as the central skill.

This isn't an argument for less medical knowledge; the editor can't catch the fabricated dose without it, and the systems thinker can't spot the unsafe default without knowing what safe looks like. Knowledge stops being the product and becomes the substrate — necessary, no longer sufficient, and no longer the thing that differentiates a good doctor from a database. The differentiators are moving to exactly the places the curriculum currently treats as electives.

What this means

Professions rarely get to choose the moment their skill distribution shifts; they only choose whether to notice. Medicine's last such moment — the great information-systems migration — was mostly endured rather than shaped, and the profession has spent twenty years documenting the cost of its absence from the design table. This shift is bigger, because the machines now draft the words and shape the decisions, not just store them. The doctors who thrive in it won't be the ones who resist the editor and systems roles as distractions from 'real medicine', nor the ones who dissolve into them and stop being clinicians. They'll be the ones who hold all three: clinical judgment as the core, editorial vigilance as its protection, and systems thinking as its reach. The title on the badge will still say doctor. The job description underneath is being rewritten now — with or without the profession's handwriting on it.