At some point in the last two years — perhaps reading a benchmark result, perhaps watching a scribe draft a better note than you would have written at 6pm — you have probably asked the quiet question: what does this mean for my job? This lesson takes that question seriously, without reassurance-by-cliché.
The stock answers are unsatisfying. “AI will never replace doctors” is faith, not analysis. “AI won’t replace you, but a doctor using AI will” is a conference slogan. You deserve the grown-up version: which parts of the GP role are genuinely exposed to this technology, which are genuinely defended, and what that means for how you spend the next decade.
What AI is already absorbing
Be clear-eyed about the exposed ground, because you are standing on some of it.
The RCGP’s April 2026 workload report found GPs reporting around a quarter of their time going on administrative tasks. Documentation, correspondence, coding, report drafting — this is exactly the work Modules 4 and 5 showed AI absorbing, and it will keep absorbing more of it. That is the point. Nobody entered medicine for the typing.
Some genuinely clinical tasks are also exposed: first-draft triage of routine demand, protocol-driven monitoring of stable long-term conditions, information-giving of the kind that fills so many consultations. Exposed does not mean eliminated — it means AI increasingly does the first pass, and the clinician’s work shifts to verification, exception-handling, and the cases that do not fit the pattern.
There is an honest discomfort here: a role built increasingly on reviewing AI output is different work from a role built on doing the task. Some of it is better — less drudgery. Some of it is harder — sustained vigilance over mostly-correct output is genuinely difficult, as every lesson on automation complacency in this course has warned. Pretending the role will not change is not a plan.
What is structurally defended
Now the defended ground — and the reasons are structural, not sentimental.
Accountability. Someone must be answerable to the GMC, to the courts, and to the patient. No AI supplier accepts clinical liability, and nothing in UK law or the MHRA’s emerging framework points that way. A system that cannot be struck off cannot hold responsibility; the buck stops with a registered professional, which means a professional must genuinely be in the loop.
Undifferentiated complexity. AI performs best on well-posed problems. General practice specialises in ill-posed ones: the symptom that is also a social situation, the presentation that only makes sense if you knew the patient’s mother, the four interacting problems that no benchmark case ever combines. The Microsoft benchmark from Lesson 2 solved exotic, well-documented cases — not “tired all the time” in a ten-minute slot.
The relationship itself. Continuity of care measurably changes outcomes — and continuity is not information transfer. It is trust built over years, risk held jointly, the licence to say the hard thing because of who is saying it. Patients will use AI for information, increasingly and legitimately. What they seek from you is judgement they can hold accountable, delivered by someone who knows them.
Examination and presence. Medicine still involves bodies. Hands, eyes, smell, the gestalt of the unwell child — and presence at the bedside, in the home visit, at the end of life. This is not a residual category. It is where much of the highest-stakes general practice happens.
The realistic middle path
Put the exposed and the defended together, and the realistic picture for the next decade looks like this.
The GP role concentrates towards its defended core: complexity, accountability, relationships, examination, risk. The routine periphery — documentation, admin, routine information transfer, first-pass triage — is progressively absorbed by tools, with the clinician verifying rather than producing. The demand side does not shrink: an ageing, multimorbid population and an access-hungry system will consume every efficiency gained, as they have with every previous productivity improvement in general practice.
Two role changes follow that are worth preparing for deliberately. First, GPs become supervisors of systems as well as of registrars — the audit, verification, and governance skills from Modules 4 and 5 stop being optional extras and become core professional competencies. Second, the profession needs GPs who can evaluate, procure, and lead on these tools; every practice needs at least one partner who genuinely understands this ground. You have just spent six modules becoming that partner.
For those who train the next generation
A closing thought for trainers, because registrars raise this constantly and deserve better than a shrug.
The registrar who qualifies in the next few years will spend a forty-year career alongside this technology. What they need from training is not protection from AI, and not uncritical enthusiasm — it is exactly what the defended ground implies: deep clinical reasoning that does not lean on the tool, the verification habits to use the tool safely, and a firm grasp of where accountability sits. A registrar who cannot diagnose without AI has not been trained; a registrar who refuses to learn the tools is being trained for a service that will not exist.
The honest summary: this technology changes what GPs spend time on, and it does so faster than the profession’s guidance is keeping up. It does not remove the need for GPs — the structural defences of accountability, complexity, relationship, and presence are real. The GPs who do best in the next decade will be those who let the tools take the typing and deliberately reinvest the time in the defended core. That is a choice, and it is yours to make.
Key Takeaway
Be honest about both halves: AI is genuinely absorbing the administrative quarter of GP work and the routine, well-posed fraction of clinical work — while accountability, undifferentiated complexity, continuity of relationship, and physical presence are structurally defended, not sentimentally defended. The role shifts from producing routine output to verifying it, and towards the defended core. Prepare deliberately: verification skills, governance skills, and the choice to reinvest saved time in the work only you can do.