.svg)
June 29, 2026
Subscribe on: RSS | SPOTIFY | APPLE PODCAST | GOOGLE | BREAKER | ANCHOR
Josh and Alan reunite for a “solo” episode to recap their adventures at AMDIS 2026 in Ojai, CA. They dive deep on how health systems are moving from AI Anxiety to AI Ops, What the EHR Might Look in 2030, the Uber That Nearly Escaped Them at the Spa, and more... Click the play button to listen or read the show notes below.
Audio:
Guest(s):
- Joshua Liu, MD (@joshuapliu), Co-founder & CEO at SeamlessMD
Episode 232 - Show Notes:
[00:00:31] What AMDIS is and why it anchors the informatics community. AMDIS — the Association of Medical Directors of Information Systems — is one of the longest-running associations for informaticists and CMIOs, founded in 1997 (nearly 30 years), and it retains a tight-knit community feel, with attendees from the late '90s still sharing stories of how it has grown.
[00:01:19] How Josh became involved with AMDIS through the Inform subcommittee — Dr. Howard Landa, chair of the AMDIS board, invited Josh to join Inform, a subcommittee launched over a year ago with two goals: evolving AMDIS' clinical informatics curriculum, and deepening the society's advocacy with partners like HIMSS so the physician informaticist voice stays central to industry and policy conversations.
[00:02:16] The Ojai setting and why a single venue changed the conference dynamic — Held about two hours northwest of LA at The Farmhouse, part of the Ojai Valley Inn (a resort known for celebrity buyouts), the remote, beautiful location concentrated top informaticists in one place. Alan and Josh booked late and stayed offsite at the Gecko Cottage, but found the intimacy beat the usual sprawling conferences where sessions span a mile-wide radius.
[00:08:17] How Nabla's sponsored receptions kept the evenings lively — Nabla sponsored both the day-one and day-two end-of-day receptions. On day two, sessions ended early but the reception couldn't begin until its scheduled time — the bartenders cited what was likely a liquor-license window — yet turnout stayed strong throughout the event's cocktail hours.
[00:13:21] Why cybersecurity is a precondition for patient safety — Christian Dameff of UCSD gave a provocative session on cybersecurity and AI from a CMIO perspective. His framing: cybersecurity is not a solvable problem but a "forever game of cat and mouse," now accelerated as AI both discovers and patches exploits faster. He argued security is a necessary precondition to patient safety — if you can't confirm your systems are secure, you can't guarantee AI outputs reaching patients are safe — and flagged how easily internet-connected medical devices like pacemakers can be exploited. He also offered the spicy take that AI could one day come for physicians' jobs, though not within the next ten years.
[00:16:34] What the emergence of a Chief Financial Informatics Officer signals — Alan and Josh heard from Dr. Carla Haack of Emory, possibly the only person currently holding the CFIO title. A surgeon who stepped back from the OR, she began helping with revenue cycle, became a translator bridging clinical IT and rev cycle, and had such an impact — saving the system millions in inefficiencies — that the role was formalized. It echoed Dr. Alistair Erskine's earlier podcast prediction about the rise of the financial informatics officer.
[00:17:53] Why the EHR of the future must be predictive and personalized — Dr. Deepti Pandita described a 2030 EHR that shifts from a "static filing cabinet" to a true intelligent care delivery system synthesizing everything about a patient. It must deliver on two Ps — predictive and personalized — with an interface that rearranges itself around each clinician and the patient in front of them. Alan and Josh compared it to Claude Skills: as a bridge, each user might maintain a markdown-style file defining which windows and views load first when the EHR boots up.
[00:22:10] How AI scribes may reduce burnout by cutting cognitive burden, not just time — Dr. Margaret Lozovatsky, newly CDIO at Premier Health, questioned whether time savings — the convenient metric everyone uses to measure AI scribe success — was ever the right one. The bigger driver of burnout relief may be reduced cognitive burden. The hosts discussed the common "it doesn't sound like me" complaint, and how a well-built, personalized prompt or skill can emulate a clinician's tone, with a HIPAA-compliant enterprise LLM as an interim workaround until personalization is native to the EHR.
[00:26:26] Why the best inbox strategy is preventing the message entirely — Dr. Marlene Millen, CMIO at UCSD, spoke on "inboxology," arguing the future of the clinician inbox is giving patients answers before they ever need to message. Rather than just drafting better replies, triage agents could handle safe, non-clinical tasks and even automate feedback on wearable data — solving the problem upstream. As Alan put it: "What's better than low cognitive burden? No cognitive burden."
[00:27:57] Why "health" versus "medical" in a title actually matters — Dr. Lacy Knight argued that CHIO versus CMIO isn't about seniority but framing: "medical" inherently centers the conversation on physicians, while "health" is role-agnostic and enables more inclusive, enterprise-wide conversations. He debated Dr. Margaret Lozovatsky, who held that titles don't matter — a rare on-stage disagreement both hosts wished happened more often, since panelist debate is where the messiness lives and the most insight surfaces.
[00:30:35] How the podcast's own community showed up across the conference — A recurring theme was reconnecting with past guests and meeting new CMIOs, which the hosts treat as a marker of a strong event. Shout-outs spanned HIMSS's Brooke Bernot Han (on a new AMDIS–HIMSS online resource for informaticists), Regard's David Kirk and Francisco Alvarez, and new connections like UCSF's Dr. Russ Cucina and Alameda Health System's Dr. Sarah Rahman.
[00:36:52] Why you should bring the AI to the data, not the data to the AI — Dr. Keith Morse, who studied computer science before medicine, argued that clinical workflows are always anchored to their original data sources, so AI should be brought to the data rather than data moved into AI tools. Keeping data in one place matters because AI solutions may change over time, while the underlying data it's built on should persist.
[00:38:23] How every new AI feature quietly adds training burden — Dr. Dave Chestek raised the crisis of an overwhelming influx of AI features, noting that each addition to the system requires more and more end-user training, compounding burden across the organization.
[00:39:29] Why patient education must match health literacy, not just language — Dr. Richard Milani's session on chronic care and patient monitoring drove home that education delivered above a patient's health literacy level is as useless as delivering it in the wrong language — the patient derives the same negligible benefit. AI tools are well suited to translating both language and literacy level, though the hosts stressed keeping humans in the loop.
[00:40:44] What a phased framework for rolling out AI tools could look like — Dr. Alistair Erskine, now leading digital and IT at Highmark (which also manages Allegheny Health Network), argued the legacy IT intake process fails for AI because it was built to assess deterministic tools, whereas AI outputs are non-deterministic. He proposed a staged rollout: no data, then synthetic data, then historical data, and finally prospective live patient data.
[00:41:39] Why AI go-lives can't simply be marked "complete" — Dr. Amy Sitapati warned that AI is often treated like any other IT implementation — go live, close the project — leaving dashboards and alerts flashing with no one watching. Her concern is invisible model drift over time when no one repeats the validation checks after go-live. The hosts agreed monitoring can be increasingly AI-assisted, but there will still need to be humans to act on the red flags it surfaces.
[00:43:34] Whether we're measuring the right outcomes for in-basket AI — Dr. Everett Weiss of Rochester Regional questioned whether tools automating in-basket draft responses are being judged on the right metrics. Beyond clinician-centric measures like cognitive burden and time saved, he asked whether longer AI-generated drafts that clinicians dislike might actually be more useful for patients — and whether anyone is studying the impact on patient outcomes. The hosts closed on the reminder that the ultimate ROI is patient care.
The Digital Patient has been recognized as Feedspot's #1 Patient Engagement Podcast of 2025. Thank you to our listeners for making this happen!
.png)
