The Forward-Deployed Physician

Why AI-native medicine needs clinicians who can distinguish what is essential from what is merely inherited. AI-native medicine needs forward-deployed physicians who can bring clinical judgment, workflow reality, and longevity medicine context into healthcare AI deployment. Medical AI cannot simply accelerate existing bureaucracy; the deeper opportunity is redesigning care around abundant intelligence. Forward-deployed physicians bring clinical judgment to AI deployment, especially at the messy edge where models meet real patients and workflows. GCLS.ai's AI and Healthcare Certification is built to give clinicians the language, frameworks, and implementation instincts to lead this transition. Why AI-native medicine needs controlled demolition When electricity first entered factories, it did not immediately transform industry. Many early factories used electric motors as a cleaner substitute for steam power while preserving the old layout of the steam age. A single large motor would drive the same shafts, belts, pulleys, and centralized mechanical architecture that had organized factory work for decades. The energy source had changed, but the operating model had not. The deeper transformation came later, when factories were redesigned around what electricity made possible. Power no longer had to be distributed from one central source through a rigid mechanical system. Motors could be placed closer to individual machines. Factory layouts could become more flexible. Production could be redesigned around a new assumption: power was no longer scarce in the same way. The [Exponential View essay](https://www.exponentialview.co/p/why-ai-isnt-showing-up-on-your-bottom-line) on AI's productivity puzzle frames this clearly through the factory electrification analogy. The early use of electric power improved the old system, but the larger gains came when factories were reorganized around distributed power. The lesson is that the technology and the operating model had to evolve together. That is the moment medicine is approaching with AI. The first phase of medical AI is attaching intelligence to the old machinery. It drafts the note. It summarizes the chart. It manages the inbox. It helps with billing, coding, and prior authorization. These uses matter, but they are still wrapped around the existing structure of care. If AI only makes the current system less painful, then we will have missed the larger opportunity. We will have built faster bureaucracy. That transition will not happen through model capability alone. It will require physicians who understand both clinical reality and AI-native implementation. I write this as a physician, and I am convinced the work needs a particular kind of clinician: the forward-deployed physician, close enough to deployment to distinguish what is essential from what is merely inherited. This has happened before The forward-deployed physician may sound new, but the pattern is old. New technologies rarely transform complex industries simply by appearing. They become transformative when people with deep domain knowledge move close enough to the technology to redesign the work itself. Toyota's production system offers a useful lens: [genchi genbutsu](https://www.lean.org/lexicon-terms/genchi-genbutsu/), often translated as actual place, actual thing or go and see. The principle is that important operational problems cannot be fully understood from a distance. Leaders and builders must go to the place where work actually happens and observe the real condition directly. Clinical AI needs the same discipline Medicine cannot be designed only from conference rooms, benchmark datasets, and product roadmaps. It has its own shop floor, and I have spent my career on it: the exam room, the ward, the emergency department, the inbox, the discharge, the prior authorization queue, and the moment of uncertainty when a clinician has to decide what matters now. Modern AI companies are rediscovering this pattern. [Palantir](https://www.palantir.com/docs/foundry/architecture-center/overview/) has long described its products as being shaped through forward-deployed engineering, with engineers working close to difficult customer problems and feeding field learning back into the platform. [OpenAI's 2026 launch of the OpenAI Deployment Company](https://openai.com/index/openai-launches-the-deployment-company/) made the same point explicitly: frontier AI deployment increasingly requires engineers embedded inside organizations working on complex, real-world problems. [Anthropic](https://www.anthropic.com/careers/jobs/4985877008) has also moved in this direction through enterprise deployment roles and forward-deployed engineers who work with strategic customers to build AI applications around real workflows. The same idea is showing up in broader industry conversations. [Aaron Levie](https://www.linkedin.com/posts/boxaaron_forward-deployed-engineers-or-roles-that-activity-7460334908486361088-MnU-/), the CEO of Box, recently argued that forward-deployed engineers, or roles with a similar motion, are becoming one of the most important functions for AI rollouts. His point was not simply that enterprises need help installing software. It was that deploying AI agents is closer to deploying work itself: understanding the business process, choosing models, building evaluations, supporting workflow change, preparing data, and tuning the agentic system against the customer's reality. [Dharmesh Shah](https://www.linkedin.com/posts/dharmesh_fdes-or-forward-deployed-engineers-are-activity-7455664967913205761-NUNo/), the founder and CTO of HubSpot, has framed the shift even more broadly. FDEs do not need to be limited to engineers; they can be Forward Deployed Experts. What matters is not the title. What matters is having people who are deep in the work that must be transformed. The frontier is no longer only the model. It is deployment, and in medicine, deployment is clinical. The polite version is too small There is a polite version of this argument that says physicians should help AI companies understand clinical workflows. That is true, but it is too small. The larger argument is that AI should force medicine to re-examine its operating model. Not cosmetically. Not by adding an AI assistant to every broken workflow. Not by asking a model to write the same note, route the same message, satisfy the same prior authorization queue, and accelerate the same institutional choreography. The question I keep coming back to is this: if intelligence were abundant, would we design medicine this way at all? This is where the forward-deployed physician becomes different from the physician advisor. The physician advisor improves the product. The forward-deployed physician challenges the assumptions underneath the product. The role requires a controlled kind of demolition: knowing which ground can be torn up, and which cannot. From forward-deployed engineers to forward-deployed physicians A forward-deployed engineer does not build software in isolation and then hand it to a customer. They sit close to the field. They understand the customer's environment, constraints, workflows, and failure points. They translate messy operational reality into working software. Medicine now needs a clinical equivalent. The forward-deployed physician sits at the edge where AI systems encounter patients, clinicians, institutions, regulations, payment models, and clinical uncertainty. Their role is to translate the real practice of medicine into systems that can actually function inside healthcare. This is different from traditional clinical informatics. It is also different from being a medical director, a physician founder, or a subject matter expert. Those roles can overlap, but the forward-deployed physician is defined by proximity to deployment: how the tool behaves, where it fits, how it fails, when it escalates, what it should not do, and how it adapts to the edge cases that only become visible in the field. The edge is where the value is AI systems are often most impressive near the center of the distribution. They summarize common scenarios, draft standard responses, identify familiar patterns, and automate repetitive work. But medicine is not only the center of the distribution. Much of the value of clinical expertise appears at the edges. The patient whose symptoms do not fit the protocol. The lab value that is technically normal but clinically concerning. The recommendation that is guideline-concordant but wrong for this specific person. The discharge plan that looks reasonable in the chart but will fail at home. The patient message that sounds safe but creates false reassurance. This is not theoretical. I have read portal messages like this one: a patient writes that their chest discomfort is probably just reflux. The language is casual. The patient is young. An AI model has seen thousands of benign messages like it and will reach for the reassuring reply. But the details that change everything are the ones I have been trained to catch: exertional symptoms, a family history, a recent stimulant prescription, a turn of phrase that quietly minimizes risk. The right answer is not a smoother message. It is escalation. The forward-deployed physician brings those edge insights back into the system. They convert lived clinical judgment into prompts, evaluations, workflows, guardrails, escalation logic, product design, documentation structures, and safety boundaries. Why natural language changes the physician's role One reason this role matters now is that the nature of software development is changing. Historically, physicians could describe clinical requirements, but they were often separated from the construction layer. A physician might say, this is what the workflow needs to do, and then engineers, designers, and product managers would translate that into software. Important clinical nuance could be lost in that translation. Large language models are changing that relationship. As more of the development process becomes mediated through natural language, physicians can move closer to the system itself. They can inspect model behavior, interrogate assumptions, shape prompts and evaluations, and ask why the system responded a certain way, what context it used, what it ignored, and how it should behave when the case becomes ambiguous. A physician can now say: This recommendation is technically correct but clinically unsafe. This summary misses the finding that actually changes management. This workflow assumes the patient has access to follow-up, but many will not. This triage logic works for primary care but would be dangerous in the emergency department. This escalation threshold is too late for this condition. Clinical judgment translated into system behavior Those statements are not generic feedback. They are clinical judgment translated into system behavior. That is the power of the forward-deployed physician: they do not need to build the model from scratch, but they need to understand the system well enough to shape how it behaves in the domain they know best. Physicians do not need to leave medicine to learn AI This is the question I am asked most often by other physicians: Do I need to learn AI? My answer is not to leave medicine as quickly as possible and join an AI company. It is to become excellent at your domain, and AI-literate enough to know how that domain should be transformed. There is a growing pattern of physicians, medical students, and recent graduates leaving the clinical path to work in AI companies. Some of this is understandable. AI feels like the future. Healthcare feels slow, burdened, and resistant to change. The opportunity to build something new is attractive. But there is a risk in exiting clinical practice too early. There is a difference between being medically credentialed and being clinically seasoned. A person can know medical terminology, understand pathophysiology, and even have a medical degree without having lived through the practice of clinical management. Residency, call, handoffs, diagnostic uncertainty, patient deterioration, family conversations, discharge planning, medication reconciliation, insurance constraints, and longitudinal follow-up all teach forms of knowledge that are difficult to acquire from the outside. The hardest parts of medicine are often not the facts. They are judgment under uncertainty, prioritization, knowing when a normal result is not reassuring, understanding how plans fail after discharge, and recognizing when a recommendation is medically correct but operationally unrealistic. An AI company does not need physicians who merely speak medical language. It needs physicians who understand real medicine. Applying this to longevity and precision medicine The same logic applies to longevity and precision medicine. For years, medicine has spoken about prevention, personalization, early detection, risk stratification, and whole-person care. But the operating model has lagged behind the ambition. A clinician may want to integrate years of labs, wearable data, imaging, sleep, exercise, medications, family history, genetics, social context, and patient preferences into a coherent plan. In practice, the system is rarely designed for that level of continuous synthesis. AI could change that. It can help maintain a living model of the patient, detect subtle changes over time, personalize education, surface preventive opportunities, and make precision medicine more usable at the point of care. But this is exactly where forward-deployed physicians are needed. A model can identify patterns, but a physician must help determine which patterns are clinically meaningful, which are noise, which should trigger action, and which should be interpreted in the context of the patient's goals and risks. Longevity is not simply maximizing biomarkers. Precision medicine is not simply matching a patient to a molecular pathway. Both require judgment about tradeoffs, uncertainty, goals, burden, risk, equity, and what matters to the person in front of the clinician. The new scarcity is domain judgment As AI makes building easier, the bottleneck shifts. The question becomes less, can we build this? and more, do we understand what should happen? In medicine, a system can be technically impressive and clinically wrong. It can be accurate in a narrow benchmark and unsafe in practice. It can produce fluent explanations while missing the operational context that determines whether a care plan will succeed. The scarce resource is not data, compute, or model access. It is domain judgment: earned through contact with patients, teams, uncertainty, consequences, and systems. The forward-deployed physician carries that judgment into the build, not to slow innovation down, but to make it survive contact with real patients. What physicians should learn So, do physicians need to learn AI? Yes, but not in the way many people think. The physician of the AI era needs both deep clinical expertise and enough AI literacy to understand how the technology changes the surface area of their domain. The forward-deployed physician should understand: what language models are good at, where they fail, how retrieval works, how prompts shape behavior, how evaluations are built, how human review loops function, how automation changes risk, and how AI systems should be monitored after deployment. This is not a call for every physician to become an AI engineer. It is a call for physicians to become more intentional stewards of how AI enters their domain. The more natural-language-native software becomes, the more physicians can participate in shaping it. But their leverage comes from knowing medicine first. That is the practical goal of the AI and Healthcare Certification we are building at DrVibe in partnership with the Geneva College of Longevity Science ([GCLS.ai](https://gcls.ai)). It is designed as an agentic learning platform, not just a place to learn AI, but a place to learn agentic design for medicine itself. The aim is not to pull clinicians away from medicine, but to give them the language, frameworks, and implementation instincts to help AI enter it correctly, and the judgment to lead its controlled demolition and rebuild. The role that medicine is missing The forward-deployed physician is not just a new job title. It is a missing institutional role. Medicine has clinicians who deliver care, informaticists who manage systems, executives who oversee strategy, medical directors who advise products and operations, and AI researchers who study models. The coming era requires physicians who can live at the intersection of clinical reality and AI deployment. These physicians will not simply ask, how can AI help doctors do what they already do? They will ask, what should medicine become now that intelligent systems can participate in care? The forward-deployed physician is the bridge: carrying clinical judgment out to the edge, carrying the edge cases back into the system, holding the line between what can be rebuilt and what must remain. That is how medicine moves from AI-assisted bureaucracy to AI-native care. That may be one of the most important physician roles of the next decade. Not because physicians need to become less clinical. Because medicine will need its best clinical minds closer to the machine. Sources Exponential View: [Why AI isn't showing up on your bottom line](https://www.exponentialview.co/p/why-ai-isnt-showing-up-on-your-bottom-line) (2026) - factory electrification analogy and productive AI deployment. Lean Enterprise Institute: [Genchi Genbutsu](https://www.lean.org/lexicon-terms/genchi-genbutsu/) - Toyota's principle of going to the source to observe real conditions. Palantir: [Forward Deployed Engineering](https://www.palantir.com/docs/foundry/architecture-center/overview/) - engineering methodology built around embedded field deployment. OpenAI: [OpenAI Launches the Deployment Company](https://openai.com/index/openai-launches-the-deployment-company/) (May 2026) - embedding Forward Deployed Engineers into organizations for complex AI deployment problems. Anthropic: [Forward Deployed Engineer roles](https://www.anthropic.com/careers/jobs/4985877008) - Applied AI team roles for embedding with strategic customers. Aaron Levie: [LinkedIn post on forward-deployed engineering](https://www.linkedin.com/posts/boxaaron_forward-deployed-engineers-or-roles-that-activity-7460334908486361088-MnU-/), CEO of Box, on why forward-deployed engineers are among the most important functions for AI rollouts. Dharmesh Shah: [LinkedIn post on Forward Deployed Experts](https://www.linkedin.com/posts/dharmesh_fdes-or-forward-deployed-engineers-are-activity-7455664967913205761-NUNo/), CTO of HubSpot, on how the role extends beyond engineers to anyone deep in the work that must be transformed. What is a forward-deployed physician? A forward-deployed physician is a clinician who works close to AI deployment, translating real clinical judgment, workflow constraints, edge cases, safety concerns, and patient context into system behavior. Why does AI-native medicine need physicians close to deployment? Healthcare AI can be technically impressive and still clinically wrong. Physicians close to deployment help decide what should be automated, what should be escalated, and what should never be redesigned away. How does this connect to the GCLS.ai AI and Healthcare Certification? The certification is designed to help clinicians build AI literacy, workflow judgment, and implementation instincts so they can guide AI into medicine safely and practically. How does this apply to longevity and precision medicine? Longevity and precision medicine require continuous synthesis of labs, imaging, wearables, genetics, goals, risks, and context. AI can help, but physicians must determine which patterns are clinically meaningful.