There's been a lot of AI news this week. Doctronic announced a new autonomous doctor granted provisional study approval by the state of Utah to pilot an AI auto-prescriber with physician oversight for a select formulary of approved pharmaceuticals, excluding controlled substances. OpenAI released a usage health report discussing how many individuals use ChatGPT to search for healthcare-related topics. And OpenAI launched ChatGPT Health, a product that leverages b.well's PHI integration to help patients use their health data with OpenAI's large language model.
All three announcements show where consumer-focused AI is heading. But is it really what's best for patients or what they're asking for?
According to a recent study in Sweden exploring primary care patients' perspectives on artificial intelligence, patients report the opposite. They state a clear preference for physician-assisted AI versus autonomous AI where no physicians or other clinicians are directly involved in the care decision-making process. Patients were quoted stating, "Although some acknowledge AI's superior accuracy, they still prefer human involvement in treatment. Many patients trust their physician more than AI and doubt that AI can replace physicians."
This is just one example of what patients are saying while tech companies choose to ignore it, powering through with their push to get AI to practice with little to no physician oversight. It's very clear what patients want: primarily access and knowledge to their specific health questions based on their data. Resources like ChatGPT and other LLM services are incredible for providing value, but if we're only listening to half of what patients are saying, we're missing the forest for the trees.
How These Services Actually Work
Let's review how these services actually work in real life versus how they're being marketed to patients. Doctronic, for example, is a model that employs doctors and clinicians to monitor and review AI to operate semi-autonomously via a ChatGPT-like interface that's actually competitive to ChatGPT Health. Doctronic's angle is that ChatGPT is unable to manage patient care and is purely informational, whereas their model allows patients to not only receive the benefits of LLMs but also access direct care services from licensed doctors and clinicians.
With their new pilot in Utah, they're able to have their AI auto-prescribe patient refills without any clinical oversight as long as there is an established relationship and care management plan in place prior to the prescription being drafted and approved for refills. Doctronic reports that they have developed their own internal monitoring, governance, and audit system for tracking these prescription refills with a custom malpractice policy to help manage any associated risk.
OpenAI's ChatGPT Health is using a third-party data integration service known as b.well in order to help them connect patient-reported data pulled directly from wearable devices, Apple Health, and some affiliated EHRs that provide data within their patient portal. Neither company has any direct proprietary EHR system of record or any internal data interoperability tool that allows them to capture data from health systems, physicians or clinicians, or any other first or second-party PHI data beyond the user-generated data taken from plug-in tools like b.well.
This is critical in determining how legacy EHR systems were able to build a true moat with their data and how this walled garden approach limits any real players from ever entering the space without significant changes in either regulations, the environment where data is stored (changes with the system of record), or the protocols for how data is transmitted from all parties to include these companies.
The Real Opportunity
Now it's important to state that we are all in with AI and are in no way against the direction of how AI is evolving to help more patients become informed of their health and create more access. But Cline sees things entirely differently than how these services are being marketed and portrayed as autonomous or how tech enthusiasts are claiming that companies like OpenAI now have some sort of healthcare data moat that will lead them to be more successful than previous companies like Apple, who was not really able to scale their Apple Health product despite having over 1 billion users using their mobile devices where Apple Health is directly embedded.
We at Cline believe that two things are converging at the same time. First, consumer daily adoption of AI for healthcare information is increasing but is nowhere near Google's adoption of over 1 billion healthcare-related searches a day with an estimated nearly 3 billion users in total per day. Second, AI will become autonomous. The key question is how autonomous and who gets to play in this space?
When it comes to the consumerism of AI for healthcare, we have to be realistic that there are a lot of legacy players who will resist this change and instead try to build roadblocks or other systems to make sure they maintain relevance as this transition occurs. One of the areas we believe needs to be challenged is the entire concept of the EHR, which is set for SaaS operations where people are the primary operators of this technology.
The Future: Clinical ADEs
Instead, we believe the EHRs of the future will be Clinical ADEs. You can visit our previous blog where we explain what an ADE is and how it works for care delivery. The primary operator for ADEs will be AI or agents that are managed via LLMs or small language models for operational or clinical use. When EHRs are Clinical ADEs and agents are the primary operators, this provides a unique opportunity to finally allow all system of record data and any data pulled in real life from the point of care to be transmitted and shared to other systems like ChatGPT Health, where patients are able to receive a complete view of their entire healthcare data, not just the data that's pulled from wearables or portals.
Cline is suggesting that we create specific communication protocols for agents to talk to other agents for the specific purpose of data interoperability, similar to protocols like MCP, A2A, ACP, or AG-UI that LLM labs are creating for agents to communicate with agents for AI code development. Once these protocols are in place, it will allow doctors to use their Clinical ADEs to create their own autonomous AI, similar to Doctronic's product, to manage and coordinate their own direct care services with their patients.
This gives them full control to not only customize and build their clinical use AI, which we refer to as digital twins, but also provide specific scenario-based context based on their professional expertise and experience in order to build for clinical edge cases that often go overlooked by companies that are building for scale and general consumer use. This ensures better safety and efficacy of treatment and management with their digital twin.
Better Together
Another benefit from this is since we know more patients are adopting AI for healthcare-related searches and that the push for more autonomous AI is inevitable for medicine, having access to a digital twin allows the doctor to better communicate with patients when they come in with questions from sources such as ChatGPT Health, which could directly be integrated within their Clinical ADE patient portal, providing 10 times better data, value, and services to both the patient and their doctor.
We believe that AI will be better for doctors and patients when it's working side by side as a clinical partner for the doctor and patient. The doctor is able to build and maintain their digital twin, giving it roles and permissions to operate autonomously based on their professional preferences, their health systems, or their malpractice guidance. This is the exact same for patients but now with direct integrations and links between their own personal health AI assistants, which they're using for healthcare knowledge and guidance.
Healthcare has always been better when doctors and patients are working together. Patients prefer it, and the data proves how outcomes, costs, and general sentiment are best when there's a relationship between the two. The use of AI should be no different, and we have to build systems to prepare for these changes in technology, not outright attempt to dismiss this key component because of the frustrations that have existed from legacy systems which consolidated care and pushed tons of middlemen such as insurance companies, technology companies like EHRs, or healthcare systems in between the doctor and patient.
The Path Forward
Cline is pushing for a future where doctors are builders of Clinical ADEs and clinical use AI because we believe that this is the future that should have always existed. Not AI replacing doctors. Not tech companies deciding how medicine should be practiced. But doctors and patients working together, augmented by technology they control, building solutions that actually serve both parties.
The autonomous AI future is coming. The question is whether it will be built by companies optimizing for scale and venture returns, or by doctors optimizing for patient care and clinical outcomes. We're betting on doctors.
Ready to Build Your Digital Twin?
Cline's EHR-native ADE platform lets doctors create their own clinical use AI and digital twins. Build exactly what your patients need. Control how autonomous it operates. Maintain the doctor-patient relationship while leveraging the power of AI.
Because the future of autonomous AI in healthcare should be built by the people actually practicing medicine.