On September 23, Nabla and Second Opinion hosted the second Accelerate Summit at 1 Hotel Brooklyn Bridge. The first edition of this series was held in Los Angeles in January. The format stayed the same: a curated group of healthcare leaders spent a day in candid conversation with each other, with no sales pitches and no product demos.
The program covered four themes: world models, the changing role of health IT leadership, the regulatory gap, and AI governance. It closed with a fireside conversation on what must not change as healthcare reinvents itself with Dr. Abraham Verghese, a practicing physician at Stanford Medicine and the New York Times bestselling author of Cutting for Stone and The Covenant of Water.
Across every session, the central question was what it takes to lead in this new era: how to move quickly without outrunning trust, build governance without creating paralysis, and embrace what AI can change without losing sight of what healthcare must preserve.
The Agenda
Welcome Remarks
Brian Manning, CEO, Nabla; Delphine Groll, Co-founder & COO, Nabla
Beyond the Copilot: What World Models Could Mean for Clinical Judgment
Suchi Saria, Founder & CEO, Bayesian Health; Alex LeBrun, CEO, AMI Labs, and Chairman & Chief AI Officer, Nabla; Brian Manning, CEO, Nabla
Same Title, Different Job: Reinventing Health IT Leadership
Amy Compton-Phillips, MD, EVP & CMO, CVS Health; David Singer, CIO, LCMC Health; moderated by Christina Farr, CEO, Second Opinion
Keynote | Mapping the Gap: Health System AI Governance When the Rules Can’t Keep Up
Jared Augenstein, Senior Managing Director, Manatt, Phelps & Phillips
Whose Governance Counts? Coalitions, Contracts, and Health Systems Caught in the Middle
Brian Anderson, MD, CEO, Coalition for Health AI; Julie Brill, National Advisor, Manatt; Rebecca Markowitz, MD, CMIO, M Health Fairview
Featured Fireside | What Must Not Change: Leadership in a Time of Reinvention
Abraham Verghese, MD, Stanford, interviewed by Ed Lee, MD, CMO, Nabla
Building the Organization You Want to Become
Manning opened the day by setting the tone: candid conversation in a small setting, built for peers rather than pitches. AI is changing fast, he noted, and it is changing organizational priorities just as fast.
Groll picked up that thread. She asked leaders to keep two pictures in mind as they navigate that change: the organization they want to become, and the one they don’t. That distinction applies as much to how organizations build as to what they build. Drawing on Nabla’s “Little Blue Book” of company culture, she named three failure modes to avoid: death by bureaucracy, death by fear, and death by ego.
Those three risks framed the rest of the day. Each is a trap for health systems trying to move quickly with AI: governance that slows everything to a halt, caution that becomes an excuse to wait, and confidence that stops leaders from listening to clinicians, patients, and the evidence.
Adoption Is Still the Hardest Problem
In a conversation on world models, Saria and LeBrun explained how the next generation of AI differs from today’s large language models. LLMs are good at knowledge, but they don’t understand the physical world. World models try to learn the way animals do, from sensory input, so they can reason about cause and effect.
For healthcare, the potential is large. Care is inherently multimodal. The long-term vision is AI that watches thousands of clinical data streams in the background and flags warning signs earlier than clinicians can today, augmenting their judgment rather than replacing it.
But the most practical lesson from the session was about people, not models. Saria stressed that adoption is still the hardest part of AI, and adoption ultimately comes back to trust. It has to be designed in from day one and measured throughout.
People come to trust new technology in three ways: by seeing the value for themselves, through a colleague they trust, and by understanding how and why it works. Organizations building an AI-native culture need all three, backed by outside validation such as FDA clearance or CMS endorsement.
The panel also raised a caution. Today’s clinicians may be a “golden generation”: they know the work deeply and have AI to support them. The next generation still needs to be trained to practice when the tools fail.
Same Title, Different Job
In a panel moderated by Christina Farr of Second Opinion, Dr. Compton-Phillips and Singer described how the CIO, CMIO, and CAIO roles are changing in practice.
Their shared view was that the job is less about controlling AI and more about corralling it. Clinicians and staff are already using these tools. As one panelist put it, “The train has left the station, but it’s not out of control.” The leader’s role is to provide governance, strategy, and clarity about which problem is being solved.
The panelists described a practical model. Teams get permission to experiment in a sandbox, and each tool is judged on quality, safety, outcomes, cost, and engagement from patients and staff. Training starts with executive leadership and cascades down.
How organizations measure success came up repeatedly. In their ambient AI rollouts, leaders deliberately did not talk about seeing more patients or raising RVUs. Documentation time still fell, and clinicians got back to work they had been unable to get to. “Solve the real, tangible problem and the money will follow.”
The panel was just as clear about who should lead these decisions. For the clinician’s voice to be heard, clinicians have to lead the conversation, or, in the panel’s words, white coats need to weigh in over suits. The warning was direct: if organizations don’t recognize real improvement when they see it, they will simply automate the dysfunctional processes they already have.
AI Governance Is a People Problem, Not a Technology Problem
In his keynote, Augenstein made the case that the real risk in deploying AI is a problem of governance, operations, and culture, not technology. By his estimate, health systems are sitting on three to five years of technology “overhang,” and the challenge now is absorbing the tools they already have.
Meanwhile, adoption is outpacing the rules. Physician use of AI has more than doubled since 2023. More than 280 AI bills affecting healthcare have been introduced in the states this year, and 37 have already been signed into law. Voluntary frameworks from NIST, the Joint Commission, CHAI, NCQA, and URAC keep multiplying, but because they are voluntary, there is little consensus among them. At the federal level, deregulation is in tension with state action, which leaves multi-state organizations facing a patchwork.
His advice: treat uncertainty as a reason to build optionality, not to wait for clarity.
The panel that followed brought that down to the health system level. Dr. Anderson, Brill, and Dr. Markowitz pointed to a significant infrastructure gap: more than 90 percent of health systems deploy third-party AI, but far fewer have the testing platforms or structured monitoring to show those tools stay safe and effective after go-live.
That makes measurement critical, but what to measure depends on the technology. Ambient documentation and sepsis prediction should not be judged the same way. The goal is to establish the right evidence for each use case and continue evaluating performance after deployment.
That also requires greater transparency between vendors and health systems. Vendors often hold the performance metrics, while health systems see only clinical outcomes. Sharing both allows for more rigorous analysis and clearer proof of value.
The panel also noted that the balance of power between vendors and health systems is shifting. The era of the 20-year contract is ending, and Brill observed that sophisticated customers have always been “the great balancer.” Well-built contracts and multi-vendor strategies can provide another layer of protection where state safeguards are thin.
Ultimately, governance comes back to trust. Roughly 70 percent of the public distrusts AI, and the panel was candid that the way forward is evidence of where AI works and honesty about where it doesn’t. As one panelist put it, this is “a use-case moment, not a hype moment.”
The question for every AI deployment becomes much simpler: What are you doing, and can you prove it?
What Must Not Change
After a day spent on world models, governance and regulation, the closing conversation turned to the part of healthcare that technology is meant to serve. Dr. Verghese joined Dr. Lee to explore what must remain constant as everything else changes.
He started with a lesson from the last wave of technology. One unintended consequence of the EHR, he noted, was clinicians chained to their computers. We may now be at the start of an era in which that burden eases. The value of that shift, he argued, depends on what the time is used for.
He described the encounter between patient and clinician as a ritual, something closer to signing a covenant, and rituals can be transformative. He shared his mother’s view that a doctor who never once touched her during a visit hadn’t really done their job. Medical training too often wears down empathy as patients become illness labels, and he urged institutions to protect meaningful time between those who give care and those who receive it.
He also asked leaders to hold two ideas at once: to take AI’s risks seriously while recognizing what it has already achieved. Even gold-standard tests are not perfect, and a human still has to interpret the results and make the call. He outlined a framework for the ideal encounter: prepare with intention, listen intently and completely, agree on what matters most, connect emotionally, and explore the patient’s understanding before offering next steps. He closed by quoting Francis Peabody: “The secret of the care of the patient is in caring for the patient.”
Looking Ahead
Accelerate NYC was built around a question bigger than how to adopt AI: what does it mean to lead in an era of intelligent infrastructure?
The leadership models healthcare has relied on are being challenged by technology moving faster than regulation, new expectations from clinicians and patients, and questions of trust that can’t be answered by technology alone. Accelerate NYC created space for leaders to step outside the day-to-day and reflect on those challenges together, not just on what their organizations should become, but on the kind of leaders they want to be as this new era takes shape.
To borrow Groll’s framing, getting there means resisting bureaucracy, fear, and ego along the way. The technology will keep changing. The responsibility of leadership is to make sure the reason for using it does not: creating the conditions for clinicians to give their time and attention to the people in front of them.
