Two Voices Healthcare AI Cannot Afford to Leave Out
Healthcare AI is having its gold rush moment. Funding is pouring in, pilots are multiplying, and every week brings another startup promising to fix documentation, triage, diagnosis, or revenue cycle. But walk through enough of these companies and a familiar pattern emerges. There is a founding team of brilliant software engineers and MBAs, a deck full of clinical aspiration, and somewhere on the advisory page a handful of physician names. Many of those advisors were asked to advise "for equity" that may never vest into anything real.
This is backwards. If we want healthcare AI tools that actually improve care rather than create new burdens, we have to be deliberate about who is in the room when these tools are conceived, built, validated, and deployed. Two groups belong at that table from the beginning: physicians and patients.
Why Physicians Have to Be There From the Start
Physicians are essential to healthcare AI development. They are the people who understand, in granular detail, three things that no engineer or product manager can fully access from outside the system.
1. How care is actually delivered. There is an enormous gap between how healthcare looks on paper, with its protocols, guidelines, and tidy workflow diagrams, and how it unfolds in a clinic at 3 p.m. when the schedule is running 90 minutes behind, the EHR is glitching, and the patient in front of you has three problems competing for the eight minutes left in the visit. Physicians live in that gap. They know which "workflows" exist only in administrative imagination and which ones reflect real life.
2. What the actual problems are, and which solutions are viable. It is striking how often healthcare AI tools solve problems that clinicians do not have, while ignoring the ones that grind them down every day. A tool that shaves 30 seconds off a task done twice a week does not move the needle. A tool that meaningfully reduces inbox burden, after-hours charting, or prior authorization friction does. Physicians can tell you the difference in the first five minutes of a conversation.
3. The friction points in workflow and implementation. A new tool does not get adopted just because it works. It has to fit. It has to load fast, integrate with the EHR, not require a fifth login, not add clicks, not break when the wifi flickers, and not put the clinician in a worse medico-legal position than before. Physicians understand these constraints because they have been on the receiving end of every well-intentioned technology rollout for the last twenty years.
The "Free Advisor" Problem
Despite all of this, too many healthcare AI startups treat physicians as ornamental. The pattern is familiar. Recruit a few well-known doctors as "clinical advisors." Compensate them in equity that is statistically very unlikely to ever convert to cash. Ask for hours of their time on calls and reviews. Then make the actual product decisions in a room they were never invited to.
This approach is exploitative, and it is also bad strategy. Equity-only advisor arrangements select for physicians who can afford to work for free, which is a narrower and less representative group than the people you actually need input from. Asynchronous "advisor" engagement is also structurally incapable of catching the design choices that get baked in early and become impossible to change later. By the time the advisor sees the demo, the product is already wrong in ways that cannot be patched.
The Data: Physician Involvement Pays Off
The case for physicians at the table has empirical support, well beyond intuitive appeal.
Research on innovation has long shown that firms founded by "user innovators," meaning people building tools to solve their own problems (think a cardiologist designing a new catheter), are more likely to receive venture funding and generate higher revenues than firms started by outsiders to the field. In healthcare specifically, when regulatory action increased the friction between device firms and physician innovators, the result was a measurable decrease in those firms' rate of innovation, especially in inventions where physician knowledge was crucial.
The pattern shows up in health tech investment as well. Among the most heavily funded and most adopted digital health companies in recent years, physician-founded and physician-led ventures keep appearing at the top of the list. Abridge, the ambient clinical AI company, was co-founded by Shiv (Shivdev) Rao, a practicing UPMC cardiologist who still sees patients while running the company; Abridge has raised over $750 million, reached a valuation above $5 billion, and is now used across more than 150 health systems including Kaiser Permanente and Mayo Clinic. Viz.ai, an AI stroke-detection platform now embedded in hundreds of hospitals, was founded by neurosurgeon Chris Mansi, MD. Doximity, the largest physician network and one of the few publicly traded health tech companies, was co-founded by Nate Gross, MD, and Roon, a fast-growing newer physician knowledge network, was co-founded by neurosurgeon Rohan Ramakrishna, MD, MD. Carbon Health was co-founded by emergency physician Caesar Djavaherian, MD. Hinge Health, the digital musculoskeletal clinic that recently went public, has a clinical operation built and overseen by physician leaders. Biofourmis, a remote-monitoring and predictive analytics company, was co-founded by cardiologist Maulik Majmudar, M.D., MD, who serves as Chief Medical Officer. Zing Health, a physician-founded Medicare Advantage insurer, has raised over $140 million. The list goes on, and it spans clinical AI, value-based care, behavioral health, primary care, and specialty care. Physician David Lortscher, founder of Curology wrote about this in his book "Why Doctors Win: The Doctor's Guide to Creating or Joining a High-Impact Startup"
The mechanism is straightforward. Payers, health systems, and patients are more likely to adopt tools when physicians help build and validate them, because the resulting product is grounded in the realities of practice. Clinical credibility cannot be bolted on at the end. It has to be built in from the first whiteboard sketch.
This does not mean every CEO needs an MD. It means medical expertise has to be part of core strategy and core product decisions, with real authority, rather than being relegated to a Slack channel that the engineering team checks once a sprint.
Why Patients Belong at the Table Too
If we are serious about building AI tools that improve healthcare, we cannot stop with physicians. Patients, the people whose bodies, lives, and outcomes these tools are ultimately about, must also be included in the design process.
This idea is well established. It is a direct extension of two long-standing movements: patient-centered care and patient-engaged research, the latter championed for over a decade by the Patient-Centered Outcomes Research Institute (PCORI) and aligned organizations.
PCORI has built its entire methodology around the principle that patients are the central focus of healthcare delivery, and therefore their wants, values, and challenges must be highly prioritized in studying healthcare choices. PCORI's approach to AI specifically is explicit. The goal is to engage patients as partners throughout the entire AI life cycle, from problem formulation to design and implementation. That is a real standard, and it is one that almost no commercial healthcare AI startup is currently meeting.
Why Including Patients Increases Success
Bringing patients into AI tool development is good product strategy. It improves the resulting tools for several concrete reasons.
Patients identify problems clinicians miss. Physicians see one piece of the patient's experience: the visit, the test result, the prescription. Patients see the whole arc, including getting the appointment, traveling to it, paying for it, understanding the instructions afterward, refilling the medication, navigating insurance, and living with the consequences. Tools designed only around the clinical encounter routinely fail at the seams between encounters, because the people who live in those seams were never asked.
Patients catch usability and trust issues early. A patient-facing AI feature that clinicians and engineers think is helpful may feel paternalistic, confusing, or frankly creepy to the people on the receiving end. The earlier this is surfaced, the cheaper it is to fix.
Patient engagement improves equity. Tools designed without input from the populations they will affect tend to underperform for those populations, and sometimes actively harm them. Including patients, and especially patients from communities historically underrepresented in tech development, is one of the most reliable ways to surface bias and gaps before deployment rather than after.
There is precedent that it works. Studies of PCORI-funded projects show that engaging patients and other stakeholders led to concrete changes in project methods, outcomes, and goals, along with improvements in measurement tools and better interpretation of qualitative data. These are the same kinds of improvements healthcare AI products desperately need.
What This Looks Like in Practice
Bringing physicians and patients to the table is a posture maintained across the entire lifecycle of a tool, rather than a single gesture at one point in time.
At problem formulation, the question of what to build should be answered with physicians and patients in the room, rather than delivered to them as a fait accompli. During design and development, engagement should be compensated, structured, and recurring for both clinicians and patient partners, with no more unpaid advisory roles. Validation should happen through real-world testing in real workflows, with feedback loops that can actually change the product. Deployment should include implementation support that respects clinical and patient context, instead of a pilot dropped on a unit and abandoned. And post-deployment, the same partners should remain engaged for ongoing monitoring, because tools drift, populations change, and what worked at launch may not work in year two.
The Bottom Line
Software engineers and MBAs are essential to building healthcare AI, and many of them are deeply thoughtful about the stakes of this work. The issue arises when they are the only people in the room.
Healthcare AI tools are going to shape how care is delivered for the next several decades. The question is whether they will shape it in ways that make care better for the clinicians who deliver it and the patients who receive it, or whether they will become one more layer of well-intentioned technology that everyone has to work around.
The difference between those two futures is decided early, in the rooms where these tools are conceived. Physicians belong in that room from day one. Patients belong there too. The time to insist on that is before the product is built, well before it has been deployed and is already failing the people it was supposed to help.
If you are building a healthcare AI tool, ask yourself some simple questions. Who is at the table, and who should be?