The authors of this perspective cannot be more wrong. 5 reasons why: 1/ It is already well known that when you clearly scope a task to a defined input set, or when the input data is clearly defined with a known prevalence of disease, AI will almost always outperform clinicians. This covers tasks like interpreting free text, differential diagnosis, or prescribing treatment or management. What is much less clear are the grey areas where interpretation is open-ended, especially when there is no clear pathological signal, such as the early onset of a chronic disease. These are usually the areas where AI models underperform and where clinical judgement plays a critical role. For example, how do you integrate a person's history, symptoms and context to decide which additional clinical tests to order and ultimately establish a differential diagnosis? 2/ The perspective heavily cites AI systems such as Google's AMIE as evidence that AI is about to surpass physicians. But note how these systems are evaluated. Much of the evidence comes from simulated clinical encounters or carefully constructed cases, which may contain far less of the noise, ambiguity and incompleteness of real-world clinical care. In such clean and bounded data spaces, it is not particularly surprising that AI performs exceptionally well. The authors themselves acknowledge that most studies comparing AI and physicians are simulations of discrete cognitive medical tasks rather than analyses of real clinical interactions. 3/ More importantly, none of this gives us enough evidence to make such a sweeping conclusion about autonomous AI. Where are the randomized, prospective, real-world trials comparing AI alone vs physicians vs AI + physicians across actual clinical workflows? The perspective cites studies mainly with carefully constructed evaluations and retrospective comparisons, but very little evidence of AI operating across the messiness of real clinical care. And yet from this relatively narrow evidence base, the argument moves quite confidently towards AI-alone care becoming superior to AI-aided physicians. 4/ Somehow I feel the authors have treated medicine like a game where the winner deserves to replace the weaker entity, which in this case happens to be the human. But medicine is not a zero-sum game. Humans have never been particularly good pattern recognizers in very high-dimensional data spaces, and this is not something they should have to do alone in the first place. Their job is to provide care using whatever tools are available at their disposal, from the simplest stethoscope to a frontier LLM. It involves understanding context, uncertainty, preferences, and treating another human being with compassion, empathy and as little bias as possible. 5/ I strongly believe the future is AI + human, not AI vs human, and I think there is even a simple statistical intuition behind this. Independent judgements can cancel out each other's errors. There is a reason aggregated or crowd-sourced judgement can outperform a standalone opinion. So the more interesting question is not whether AI will replace physicians. It is how we can design AI + clinician systems where one compensates for the weaknesses of the other, and together they provide better care than either could alone