David Silverstein started building Amaze Health in 2019, before the current AI wave made “AI health” feel like its own category. He told me he was not trying to invent the models, he was trying to build the platform that would be ready when the models were ready.
Silverstein, CEO, describes the company as an operating system for healthcare, something employers and patients can use to manage care the way a computer runs apps. The product started in what most people think of as telemedicine, then expanded into areas like mental health, virtual primary care, ortho support, navigation, and care coordination, all packaged as a subscription so the user is not paying a new fee every time they need help.
Where Amaze is trying to draw a line is how it uses AI. Silverstein says the company has its own medical team and a national medical practice, with providers as full time employees. That setup lets them use AI mostly behind the scenes, not as a chatbot trying to replace a clinician, but as software that makes clinicians faster and better.
He gave a simple example. If a patient messages in about a rash, Amaze can prompt for a photo while the call is ringing, and he says they answer most calls in 15 or 20 seconds. Once the photo is captured, the AI can do visual pattern recognition and flag a likely match, poison ivy was his example. The patient can still wait for the clinician, ask for more info, or hang up if they already know what to do. In parallel, the clinician sees the photo and a probability score on their screen.
A second example gets at why this is not only about convenience. Silverstein described a scenario where a parent calls about a child vomiting overnight. While the clinician is asking the normal questions, the AI is transcribing, taking notes, and running a rolling diagnosis. He said the system could call out to CDC data through an open API, spot that there have been recent E. coli cases near the caller, and surface that context to the clinician in real time.
The line he kept coming back to was “meeting patients where they are.” Some people want a human every time. Some people want a fast answer and a safety net. Silverstein’s view is that a hybrid system can let the patient choose the pace, while still keeping a real clinician in the loop.
He also talked about public health signal. Amaze is not trying to train new medical models from scratch. He says the company is still relatively small in data terms, with a little over 100,000 patients, and he does not think it would be responsible to claim they are “improving medicine” through training at that scale. But he does think they can spot patterns faster than traditional reporting loops. He described seeing COVID flare ups in near real time because of concentrated employer use in Phoenix, then checking later and seeing the same spikes reflected in CDC reporting after a delay.
On the product side, he says they already use AI to generate provider notes and to evaluate patient sentiment, including whether a patient sounds receptive to instructions. He also mentioned using AI to score a “patient activation measure,” a way to estimate a patient’s knowledge, skill, and confidence in managing their own care, and then use that to tailor how clinicians communicate on future calls.
Silverstein’s bet is that healthcare will get more technical under the hood and more relationship driven on the surface. In his view, AI will keep getting better at complex diagnosis, but patients will still need a trusted partner to help them decide what to do next, where to go, and how to pay for it as the system gets more confusing.
