Today @Nature published 2 new AI medical agentic models that take capabilities to a new level, from end-to-end care after presentation to the emergency department and longitudinally through 3 out-patient visits. There's a lot to unpack. Summarized in a new Ground Truths post. https://erictopol.substack.com/p/agentic-ai-comes-to-medicine
Positive users hail the new AI medical agent models in Nature as a major leap for longitudinal care, while negative users criticize their reliance on eminence-based guidelines and limited real-world validity.
No Digg Deeper questions have been answered for this story yet.
Until now, we haven't seen medical AI models intended to provide end-to-end care after presentation to an emergency department or longitudinally via 3 sequential outpatient visits. There's a lot to unpack in 2 new @nature publications today. Summarized in a new post.
Links to the papers reviewed in the post above https://www.nature.com/articles/s41586-026-10675-5 https://www.nature.com/articles/s41586-026-10764-5
Thanks for these important contributions and congrats @jnkath @Dykex6 @HeidelbergU and @HardyShakerman @valentinlievin @apalepu13 @GoogleDeepMind @Google and their colleagues
@EricTopol @Nature AI is the new gold
@EricTopol @Nature Healthcare is fundamentally longitudinal. Agentic AI that can reason across time may represent a major leap beyond traditional clinical decision support.
@EricTopol @Nature the across-3-visits part is what stands out. single-visit diagnosis has been decent for a while, but holding context between visits without it drifting is the harder problem. does the eval score that handoff, or each visit on its own?
@EricTopol @Nature Very soon
@EricTopol @Nature the 'not yet representing real world medicine' is doing a lot of heavy lifting there
A atenção primária em saúde é o cenário mais desafiador para modelos de IA, mas é inegavelmente de maior alcance para saúde pública. O desafio é a miríade de informações objetivas irrelevantes para um diagnóstico definitivo coletada ocultar o sinal procurado. O fator humano aqui ainda leva vantagem, por captar aspectos subjetivos em cada contato direto, que muitas vezes guiam de modo não-probabilístico o raciocínio diagnóstico. Entendam, medicina ainda é arte, mais do que ciência.
@EricTopol @Nature This feels like a major step toward real clinical AI integration
@EricTopol @jnkath @Dykex6 @HeidelbergU @HardyShakerman @valentinlievin @apalepu13 @GoogleDeepMind @Google Precision isn't judgment. Topol's right. The dual-agent design with 85K action options is what actually ships.
@EricTopol @Nature 100% guideline adherence sounds like a feature until you remember most guidelines are eminence-based. The AI optimized the wrong function beautifully. Precision isn't correctness.
@EricTopol @Nature Zero knowledge. Zero proof stored. Zero access to user data. User-Owned Truth. We don't host your evidence. We don't host your diagnosis. We host math that seals proof in a file. http://Cliniclogic.org http://aubreylogicai.com http://Quad.Aubreylogic.com
