252: Montefiore’s CDIO Deepesh Chandra: When Can You Stop Checking an AI Agent’s Work, Why AI Agents Will Add, Not Replace, Channels, and How the CDIO Role Became Inseparable from Operations

Subscribe on: RSS | SPOTIFY | APPLE PODCAST | GOOGLE | BREAKER | ANCHOR

On this episode of The Digital Patient, Dr. Joshua Liu, Co-founder & CEO of SeamlessMD, and colleague, Alan Sardana, chat with Deepesh Chandra, Chief Digital & Information Officer at Montefiore Einstein, about "When Can You Stop Checking an AI Agent’s Work, Why AI Agents Will Add—Not Replace—Channels, How the CDIO Role Became Inseparable from Operations, and more..."

Audio:

Guest(s):

• Deepesh Chandra, Chief Digital & Information Officer at Montefiore Einstein
• Joshua Liu, MD (@joshuapliu), Co-founder & CEO at SeamlessMD

Episode 252 - Show Notes:

[00:06:06] Why layering AI on top of legacy systems is where health systems start, and where most will stall — Deepesh says the real test of weaving AI into care is whether it "can act on the same rails and same workflows that our staff uses and not just talk about them." Fully baking AI into everyday workflow will take time and change management. Until then, AI will largely act as another contextual layer, another mode of communication, and another door for consumers, patients, and physicians.

[00:07:32] Why AI agents will add a channel rather than replace existing ones — mail, email, and phone calls didn't disappear when apps and text messaging arrived, and Deepesh expects AI to function the same way for a good period of time. Agents can technically book appointments today, but before health systems let outside agents into their systems, the industry needs to understand who those agents represent and how patient information is used, retained, and processed, and must start "building a network of trusted agents."

[00:10:17] Why the cost of being wrong keeps patients from handing healthcare decisions to AI — healthcare professionals consistently rank among the most trusted, and "the cost of being wrong about getting your healthcare wrong is way higher than cost of getting your insurance wrong." Even for something as simple as picking a doctor, Deepesh says people still want validation from a primary care doctor, a trusted friend, or a family member.

[00:11:59] What an AI-native health system could look like — much of healthcare remains bedside and exam room centric, requiring a physician and patient in the same place. Deepesh sees the AI-native opportunity in everything else, from wellness and preventive care to early, lower-acuity triage. Health systems will still need to put their "seal of approval" on these services by bringing trusted clinical guidelines and protocols into AI models, with fail-safes and guardrails so AI "is not an open chat box." He calls this a near-term reality.

[00:17:12] How AI can open up care that capacity never allowed, starting with loneliness — loneliness is a growing problem among older patients and a known accelerant to other health conditions. Deepesh describes an AI agent that can patiently call an elderly patient and hold an hour-long social conversation at minimal cost. Wherever there is a lack of access, capacity, or efficient process, AI-native capabilities have room to help, as long as trust, governance, and medical guidelines are addressed.

[00:19:02] Why post-discharge follow-up calls are a prime AI use case — hospitals and practices don't have the capacity to call every patient after a specialist visit or discharge, so they focus on those who need it most. For the longer tail, an AI could call on the hospital's behalf, check on new medications and wound healing, and immediately flag red flags for medical staff. Deepesh calls it a satisfaction tool for patients, hospitals, and the broader ecosystem, limited today only by capacity.

[00:22:21] How a three-layer model clarifies who owns patient data when AI agents get involved — at the bottom is the medical record, where the health system is the custodian and patients can share data via API with consent. At the top is the agent layer that executes tasks. In between sits the "harness" of context and knowledge: socioeconomic information not in the record, medical protocols and guidelines, and rules such as which mental health topics should trigger a handoff to human staff. That middle layer can be jointly owned by the patient, health system, and agentic product company. "If anyone tries to take control of all the three layers, I think this is where the red flag is."

[00:26:34] Why liability becomes the defining question as AI moves toward autonomy — when AI assists a human, liability shifts toward the human or institution. In fully autonomous workflows, Deepesh points to autonomous vehicles, where liability sits with the company running the vehicle. Healthcare will need to define where liability sits, and whether AI companies are financially viable enough to carry it across the volume of patients they may touch.

[00:28:20] Why governance is the brakes that let healthcare move fast with AI — "We can go as fast as AI can take us, as long as we have the right brakes on this car." Because AI produces output faster than humans can validate it, governance has to establish trust in that output and in how data is processed and stored. Deepesh flags unresolved questions about whether data becomes part of a model's "latent memory," noting that vendors often say yes, pause, and then say "I will have to get back to you on that," a clear risk for companies building wraparound applications on foundation models.

[00:31:41] How long it takes to trust an AI agent like a new hire — responding to the idea of inspecting an agent's work until it proves reliable, Deepesh says trust will take years, just as it does with people, and every negative outcome "chips away" at it in a "sinusoidal" pattern. Technology is being created faster than it can be adopted: the iPhone took nearly two decades to earn its place, yet "we are trying to trust AI within three years." He draws the line based on "the cost of going wrong" and how reversible a decision is.

[00:34:38] How Montefiore knows its digital front door is expanding access, not just serving the already-connected — Deepesh sees technology as a powerful tool for democratizing access, lowering language, access, and ability-to-pay barriers. A patient who can't make a fifteen-minute call at 10 a.m. can book on their phone on the bus at 8:15. With a payer mix that is 80 to 85% government (Medicare, Medicaid, and duals), significant upticks in digital appointments, pre-check-in, and digital payments show that a disproportionate share of adoption is coming from underserved patients.

[00:36:52] Why digital adoption still depends on manual, in-person education — adoption is slow, and often patients "just never knew that we had these capabilities." Montefiore distributes QR codes and trains front office staff to remind patients at every visit: "Next time try the app. Next time try to text this thing."

[00:37:57] How a patient ambassador program drove ambient listening adoption — physicians were quick to adopt ambient listening, but patients needed to understand that AI would be listening before walking into the exam room. Deepesh frames it as a consent management, practice transformation, and patient trust issue that requires a human in the moment. Montefiore saw a significant uptick in adoption when an ambassador was in or before the exam room, especially for patients facing a language barrier.

[00:40:06] Why the CDIO role is now entangled with operations — the role used to center on technology execution, "install, install and install," such as delivering networks and EMR adoption. Since COVID, it has become closely tied to the chief operating officer's role. A decade ago, staff knew how to run downtime procedures when the EMR failed; today, technology disruption can nearly take a hospital out of delivering services, much like a store that stops when checkout goes down. Technology now plays a role, small or leading, in virtually every major strategic initiative.

[00:44:29] Why change management now costs two to three times the technology itself — as technology gets cheaper and change management needs grow, Deepesh says the old dollar-for-dollar estimate is becoming "$2 or a $3." The patient ambassador program behind Montefiore's ambient listening scale-up wasn't a vendor recommendation and wasn't accounted for in the contract. AI in particular tends to rewire workflows the way a new EMR or ERP does, requiring process re-engineering, moving work between people and machines, and sometimes involving HR.

Fast 5 Lightning Round:

1. What is your favorite book or book you've gifted the most?

First, Break All the Rules by Marcus Buckingham and Curt Coffman

2. If you could instantly master any skill, what would it be?

"The power to influence"

3. Would you rather have super strength, super speed, or the ability to read people's minds?

"Super speed, for sure. I'm a fanatic for speed."

4. What is something in healthcare you believe that others might find insane?

"You cannot get a straight answer to who pays for the service"

5. What is the last movie or TV show you saw, and what did you think of it?

"The Pitt. Eagerly waiting for the new season in January."

The Digital Patient has been recognized as Feedspot's #1 Patient Engagement Podcast of 2025. Thank you to our listeners for making this happen!