Carlos Reis

Carlos Reis writes almost entirely about one intersection: artificial intelligence, clinical governance, and the practical work of running health institutions. Across his posts a consistent argument emerges. AI is not a threat to be resisted or a magic fix to be adopted blindly, but a force that has to be organized, supervised, and matched to human judgment at every level. He returns again and again to the same structural point: technology usually is not the bottleneck. When discussing Brazil's CFM Resolution nº 2.454, he notes that hospital leaders "já conhece a Resolução CFM nº 2.454, mas ainda busca entender como transformar suas exigências em processos dentro da instituição," and concludes plainly that "a maior dificuldade quase nunca é tecnológica" . The real work, in his view, is institutional: defining responsibilities, evaluation criteria, and monitoring, and simply "dar o primeiro passo de forma estruturada" .

This governance instinct is not abstract for him; it is calibrated to risk. He insists that "nem toda IA na saúde exige o mesmo nível de governança," distinguishing a document-organizing tool from one that "influencia decisões clínicas," and argues that monitoring and controls "devem ser proporcionais aos riscos envolvidos" . That same proportionality shows up in how he frames the evidence base of medicine itself. Citing researchers from UCL, Moorfields Eye Hospital, Tsinghua University and Singapore National Eye Centre, he lays out a hierarchy of data, information, evidence, and clinical practice, arguing that large language models can radically transform the first two levels, while the move to evidence still depends on method and validation . His summary of the field's core tension is characteristically compact: "LLMs aceleram a produção e a síntese da informação. Evidência científica continua dependendo de método e validação" .

A second recurring theme is his refusal to let efficiency substitute for critical thinking. Discussing an MIT Sloan Management article on what he calls "AI gravity," he warns that "quanto mais eficiente a IA se torna, maior a pressão para usá-la como atalho cognitivo," and frames the real risk not as using AI but as "perder a capacidade de reconhecer quando ela está errada, incompleta ou desalinhada com o contexto" . He explicitly endorses the article's prescription to use AI "como uma treinadora cognitiva, não apenas como uma fornecedora de respostas prontas" . This same conceptual clarity appears when he distinguishes generative from agentic systems, summarizing an MIT Sloan Management Review framework: "A IA generativa responde perguntas. Agentes de IA perseguem objetivos" . He is drawn to frameworks that let him separate categories cleanly, whether it is data versus evidence, or chatbots versus agents, and then apply that clarity to concrete health-system decisions.

Reis positions physicians and medical leadership, not technologists, as the rightful drivers of this transformation. He states directly that "poucas vezes na história da Medicina surgiu uma oportunidade tão relevante para que médicos liderem uma transformação profunda na forma como a assistência é prestada," arguing that clinical training and assistential experience are precisely the knowledge base needed to implement AI "de forma segura, ética e centrada no paciente" . He frames clinical governance not as a regulatory burden layered on top of medicine but, in his words, as an opportunity rather than merely "uma nova obrigação regulatória" . This leadership argument connects to his own long-running work with clinical data: he describes nearly twenty years spent turning clinical data into applied intelligence for intensive care, with a base now covering "mais de 10 milhões de pacientes críticos, provenientes de mais de 2.300 UTIs em 15 p[aíses]" , and he treats this scale as evidence that data-driven governance produces measurable results, pointing to a state-level clinical governance project in Goiás as a concrete example "gerando resultados concretos" .

On patient safety and organizational learning, he pushes back against a common managerial error: treating low incident reporting as automatically good news. Commenting on an article by Laiane Silva, he warns against "confundir ausência de evidência com evidência de ausência," noting that low notification rates can reflect excellence but can equally reveal "medo, baixa adesão ou falta de confiança no processo." For him the real marker of maturity is not the absence of reported incidents but "a capacidade da organização de aprender com os incidentes e transformar esse aprendizado em melhorias concretas" . This is consistent with his broader worldview that metrics and technology are only useful insofar as institutions build the judgment and processes to act on them, echoed in his account of an internal AI agent competition at Epimed Solutions Brazil, where he concludes that the biggest outcome was not any individual AI agent but the fact that "cada apresentação permitiu que todos conhecessem melhor o trabalho das outras áreas" .

Outside the AI and governance material, Reis occasionally steps back to offer more personal, reflective observations that still tie into decision-making. On hiring, he describes learning that first impressions of candidates are more predictive than resumes or interview performance, summarizing it as "a entrevista começa antes da primeira pergunta," and adds that whenever he ignored a negative first impression and hired anyway, he "quase sempre descobri, mais tarde, que minha primeira impressão estava certa" . On navigating crises and market turbulence, he invokes an expression he picked up early in his career, "siempre que llovió, paró," adapted into Portuguese as "sempre que choveu, parou," using it to argue that not every decision needs to be made "no auge da tempestade" and that "calma e paciência também são estratégias" . These pieces round out a picture of someone whose operating philosophy, whether applied to AI governance, hiring, or crisis management, consistently favors structure, proportionality, and patience over reflexive action.