For three days in early August, WONCA’s International Classification Committee (WICC) listserv ran its most substantive AI debate of the year — not sparked by a vendor pitch, but by an academic paper arguing AI is about to make narrow medical specialties obsolete. The family physicians who pushed back landed on a sharper, less comfortable question than the one the paper asked.

The debate opened with Belgian family physician Marc Jamoulle circulating “How AI will redefine care delivery: the rise of the generalist-specialist” (Health Affairs Scholar, 2026), by Bob Kocher, Siobhan Nolan-Mangini and UCSF hospitalist Robert Wachter — three authors from health-tech and hospital medicine, not primary care. Their claim: AI’s ability to match specialist-level knowledge across domains erodes the “cognitive necessity” for narrow specialties, freeing clinicians to become AI-augmented “generalist-specialists” managing whole disease clusters rather than single organs. Buried in their own paper, though, is a concession that undercuts the headline: current AI models are least reliable at exactly the thing family medicine already does — “context-dependent, longitudinal clinical reasoning: synthesising fragmented histories, eliciting subtle findings, and navigating patient-specific uncertainty over time.” Jamoulle seized on that line, tying it to his own long-COVID work.

Brazilian GP Gustavo Gusso countered with what he called the “specialists bias” he hears at home: not that generalists absorb specialist knowledge, but the reverse — AI itself becomes the generalist, and human specialists retreat further into their niches. Ghent’s Jan De Maeseneer, a senior WONCA voice, rejected both framings: the evidence so far shows AI helping narrow specialties (his example: radiology) while remaining “of little, if any help” in family medicine’s complexity — “for the time being.” Belgian researcher Benjamin Fauquert went further, arguing longitudinal reasoning needs no AI assist at all for a GP who already works with care episodes and problem lists — “a huge waste of energy except for some very complex cases” — and noted drily that WHO has argued against narrowly specialty-defined care for fifty years; it’s taken an AI panic for hospital medicine to arrive at the same conclusion. One skeptic on the thread, jksoler, pointed out that Jamoulle’s claims have no Belgian primary-care data to test them against, since the national “Transition Project” registry was discontinued. De Maeseneer, defending Fauquert’s line, pointed to WHO’s own move to get ahead of the argument entirely: Regional Director Hans Kluge’s 15 July Lisbon summit, where 37 countries convened on AI governance in health systems — a reminder that policymakers are already regulating around this technology while the profession is still arguing over whether the underlying risk is even real.

That’s where the thread’s real finding sits — not in the “will AI replace the generalist” framing at all, but in a document Hong Kong GP Gene Tsoi introduced partway through: Dr Benn Gooch’s account of quitting ambient AI scribing after eighteen months of enthusiastic use. Gooch’s case wasn’t that the tool failed — it worked exactly as designed — but that it quietly removed “the clinical narrative: the selective, interpretive story a GP constructs to capture not just what was said but what mattered.” That matches what the wider literature is starting to show, independent of anyone’s theory about generalist futures: a Health Affairs Forefront analysis of deskilling risk in ambient documentation reports measurable drops in unassisted diagnostic performance after sustained AI use, with trainees — who haven’t yet built the habit AI is replacing — most exposed; a separate physician survey splits deskilling concern almost evenly across three worries: automation bias and reduced vigilance, deskilling of new doctors specifically, and erosion of clinical judgement and empathy generally, each cited by roughly a fifth of respondents. None of this is one-sided: the same evidence base also shows real reductions in burnout and documentation time, and better eye contact with patients. But the WICC thread’s actual insight, arrived at almost by accident through Gusso’s closing observation that “the act of registering corresponds to the act of reasoning,” is that the threat to family medicine was never replacement. It’s what happens to a generation of GPs who never have to practise the reasoning that writing the note used to force — and that risk lands hardest exactly where postgraduate family medicine training is thinnest and most documentation-dependent as a teaching tool, which describes most African training programmes long before it describes Ghent or Brussels. Nobody on the WICC thread made that connection explicitly. It’s the one AfroPHC readers should.

Also from this week’s AfroPHC digest: