Episode 5: Digital Empathy and Why AI Scribes Are the First Technology Doctors Actually Want
Welcome to Episode 5 of the Ardexia Podcast where we speak to leaders, clinicians and advocates for better healthcare. This week, we talk to Dr. Matt Sakumoto, a virtualist primary care physician in San Francisco, fellowship-trained clinical informaticist at UCSF, regional CMIO, and my longtime collaborator on digital empathy in AI.
EHR systems and telehealth platforms are designed for billing. Patients expect warm, human digital experiences. That disconnect creates trust problems that kill adoption.
Dr. Matt Sakumoto has spent his career at the intersection of clinical care, technology, and health systems scale. As a virtualist whose entire practice is digital and a CMIO responsible for implementing efficiency tools in the EHR, he understands what makes technology actually work versus just impressive in demos.
The conversation revealed something critical: AI scribes are the first healthcare technology where doctors created waitlists instead of resistance. Understanding why tells us everything about successful digital health adoption.
Trust Is the Key to Digital Empathy
I asked Matt about the disconnect between how systems are designed versus what patients expect.
His answer: "The key to empathy is trust. And it's making sure trust and just clarity of communication."
Digital empathy isn't about finding perfect words or sophisticated technology. It's about the same things we do verbally, just in written format.
"It's the teach back or the read back: 'Just to clarify, I want to make sure I heard you correctly. Is this your plan going forward? Did you understand the plan we had for you?'"
Simple. Obvious. Critical.
"Did you hear me? Yes, we heard you. We also heard you. It's that clarity and connection that drives empathy, more so than what specific word to use."
The First AI Tool Doctors Want: Ambient Documentation
We discussed AI tools in the EHR that genuinely reduce cognitive load. Matt's answer was immediate: AI scribes.
"It has been personally really freeing to actually just look the patient in the eye and talk with them and not worry about capturing all the details, trusting that's getting captured in the background."
But here's what matters: This is the first technology where Matt had to create waitlists instead of begging clinicians to try it.
"Before I'm like, 'please just try it, please just pilot it.' Even telehealth required the pandemic. But this is the first time I've had to create wait lists for technology."
Why? "It just fits into the workflow. Doctors talk with patients, and this happens in the background. Every other technology required changing how you approach the visit. This one lets you take a step back."
At worst, neutral. For most? "Significant impact on work-life balance."
Cognitive Presence vs. Cognitive Burden
Matt described what he calls "the duck swimming underwater thing"—fingers at the keyboard typing while talking to patients in video visits.
"When I'm able to incorporate ambient documentation, there's just that level of focus I'm able to have. From a cognitive burden standpoint, that's made a significant difference."
This is what efficiency tools actually need to address: not just time savings, but cognitive presence.
Can you be fully present with your patient? Or is part of your brain tracking documentation, billing codes, what you're forgetting to capture?
AI scribes remove that background processing. You can just be a doctor.
Evidence-Based AI and the Evolving Standards
I asked how he thinks about evidence-based implementation for AI tools, especially those not used for clinical decision-making.
"How we generate evidence needs to change. With AI that's constantly learning and adapting, we have to switch from ironclad RCT approaches that take 10 years and really have monitoring and reacting in real time."
The new framework: implementation science concepts and real-time data tracking rather than decade-long studies before deployment.
This matters because startups don't have 10 years. Healthcare systems need solutions now. We need frameworks for safe deployment that don't require extensive RCTs before implementation.
Non-Inferiority vs. Perfection
I referenced Stanford's recent preprint showing large language models could produce harmful medical advice at 22% rates through errors of omission.
Matt's response: "Humans are also fallible. What is the standard we're going off of? Going for perfect is too difficult. We should go for non-inferiority rather than perfection."
His framework: Think of AI as the intern. Trust but verify.
"I guarantee I also miss some percentage of things as a human. So at least in these early stages, AI is the intern."
This is the pragmatic approach healthcare needs. Not "AI must be perfect" but "AI must be at least as good as humans, with appropriate oversight."
Virtual Care Beyond Video Visits
Matt clarified what being a "virtualist" actually means:
"People think of telehealth as video visits only. Now, a lot of virtual care is asynchronous messaging. That's where large language models can really thrive."
Example: Patient messages "I have a cough." There are the same 5-8 follow-up questions Matt always asks.
"If something can automatically do that and send it over to me, that's huge. Faster for the patient, huge cognitive burden off of me."
The clinician remains the final arbiter. But information gathering? "Never fully on autopilot just yet, but there are things I tend to automatically ask anyway, and I can see that portion be automated."
The Startup Advice: Start with the Problem
Matt advises multiple early-stage health tech companies. His guidance: "Ask questions. The whole care team. What are the problems you're trying to solve? Start with the problem. People skip that step. Defining the problem is 90% of it."
He referenced my example from Jefferson's telemedicine where we assumed parents struggled with website design for separate child charts. The actual problem? Registration workflow.
"These little things happen all the time."
You must try using it as a patient would. Not assume. Not theorize. Actually do it.
The Bottom Line for Digital Health Implementation
Three principles emerge from this conversation:
AI Scribes Succeed Because They Don't Change Workflow Every other technology forced clinicians to adapt. AI scribes adapt to clinicians. That's why doctors create waitlists instead of resistance.
Evidence Standards Must Evolve for AI Ten-year RCTs don't work for technology that learns and adapts. We need real-time monitoring frameworks and non-inferiority standards, not perfection.
Digital Empathy Is Trust and Clarity Not fancy words or sophisticated technology. Just: "Did you hear me? Yes, we heard you." Confirmation. Connection. Presence.
Matt's advice for young physicians interested in informatics or virtual care: "Seek out the community of physician entrepreneurs and innovators. It's bigger than you think."
The technology that succeeds in healthcare is the technology that fits into existing workflows, demonstrates clear value quickly, and maintains the human connection that makes medicine work.
————————————————————————————————————————————————————————
Learn more about Dr Matt Sakumoto's work and follow his insights : LinkedIn | AMA Interview | Digital Empathy 2.0
Watch the full conversation on Youtube.
Listen on Spotify or Apple Podcast.
Follow us to hear our next episode, drops every 3 weeks.
Ready to transform your healthcare innovation approach? Contact Ardexia to discuss how we can help you move from pilot to sustainable adoption.
Related Resources
Dr. Aditi Joshi is the CEO of Ardexia and host of the Ardexia podcast. She's an emergency physician who has built multiple digital health programs across three continents and specializes in turning failed digital health implementations into measurable clinical and financial success.