When health systems moved quickly to appoint chief AI officers around 2024, the role was largely theoretical. A year or two later, the first wave of AI chiefs is discovering what the job actually demands.
Over the past 18 to 24 months, health systems such as Cleveland Clinic, Los Angeles-based Cedars-Sinai and UC San Diego Health have stood up dedicated AI leadership jobs, often without clear precedents or playbooks. The mandate was ambitious: accelerate adoption of AI while ensuring patient safety, clinician trust and organizational accountability.
Now, as several of those early incumbents approach or pass the one-year mark, the position is coming into focus:
1. The job is less “AI evangelist” and more “AI skeptic-in-chief.” Early assumptions about the role often tilt toward hype: that a chief AI officer will push AI everywhere, fast. Several leaders said the reality is closer to the opposite — building credibility by not defaulting to AI.
Karandeep Singh, MD, who joined UC San Diego Health as chief health AI officer in late 2023, said a mistaken notion is that his title signals he’ll “be using AI for everything,” when his posture is more cautious and consensus-driven. “The biggest misconception … [is] the feeling that I’m going to be pushing AI to people where they don’t want AI,” he told Becker’s. “We have the ability to try something, realize it’s not working and stop.”
2. Speed is misunderstood: Transformation is cumulative, not instantaneous. If there’s a universal surprise, it’s how many stakeholders assume AI will flip the system overnight.
Mouneer Odeh, who has been chief AI and data officer of Cedars-Sinai since late 2024, pointed directly at expectation-setting as a core duty: “Real transformation is cumulative, not instantaneous.” He referenced Amara’s law — leaders overestimate short-term impact and underestimate long-term effects — arguing that AI’s benefits are real, but they compound through learning, workflow change and maturity.
3. “Data readiness” becomes the hidden governor on ambition. The flashiest demos are often the easiest part. The slow work is getting inputs reliable enough for clinical and operational trust.
Mr. Odeh said the biggest barrier hasn’t been cultural resistance so much as data readiness. “We can often build a working prototype in hours, but ensuring the underlying data is accurate and AI outputs are reliable takes far more time and attention,” he said. He gave a concrete example — training an AI agent on EHR workflows required reviewing “thousands of job aids and training materials developed over a decade.”
Alok Chaudhary, who was named vice president and chief data and AI officer of Richmond, Va.-based VCU Health in mid-2024, agreed in blunt terms: “AI is all about data. So if you don’t have the great data foundation … AI is not going anywhere.” He described spending much of his first 18 months building “foundational capabilities,” including a modern data platform, bringing in data from multiple systems and launching a formal data governance program.
4. The biggest early wins aren’t always clinical. Ben Shahshahani, PhD, chief AI officer at Cleveland Clinic, who was appointed to the role in July 2024, said one of his early surprises was how much value exists outside of clinical care, particularly in operational areas such as revenue cycle management and documentation quality.
“So much of the opportunity is in the mundane stuff in healthcare, revenue cycle management and documentation quality,” Dr. Shahshahani said. “I never thought there was actually so much opportunity there, particularly in an industry where margins are low.”
5. The role evolves into a coordinator of a matrix — because AI shows up everywhere. AI is not one program; it’s dozens of use cases across clinical, operational, research and vendor ecosystems. That pushes AI chiefs into enterprise orchestration.
“In an ideal world, our AI strategy is just embedded into our health strategy. It is not a separate strategy,” Dr. Singh said. The differentiator, he added, is knowing when leaders are talking about “one big bucket” versus “a lot of little things,” and helping them match the right kind of AI (predictive, generative, etc.) to the problem.
Mr. Odeh described relationships across clinical, operational, technology, innovation and compliance functions as central to the job, noting he’s partnering with leaders spanning clinical informatics, AI health research, ambulatory ops, revenue cycle and government affairs to keep pace with vendor and regulatory change.
6. Governance isn’t a committee — it’s the operating system for trust. By the time pilots start proliferating, governance becomes the scaffolding that keeps the organization from accidental harm, reputational risk and “shadow AI.”
Dr. Singh highlighted how governance shows up as practical, systemwide standards. One example: a “universal AI disclosure” for any AI-influenced content that reaches patients. If AI is used in documentation or messaging that patients read, he said, it needs a standardized blurb across approved tools. The reason is trust: “If we use AI in something and a patient didn’t know … and discovered it later, they might feel that as a breach of trust. And so we want to be transparent upfront.”
Mr. Chaudhary described building an AI governance body that draws from medical, information security, privacy, compliance, legal, risk, regulatory affairs and business leaders — co-chaired by his AI manager and a physician informaticist — meeting biweekly to review feasibility and risk. His takeaway: Governance maturity takes time. “We are still learning … but we are improving.”
7. Workflow fit and change management determine whether AI succeeds. Even when AI tools perform well in demos, success depends on whether they fit into real clinical and operational workflows. Adoption challenges rarely stem from the technology itself, Dr. Shahshahani said.
“It’s not just about the technology,” he said. “Eighty percent of the challenge is workflow and change management.” At Cleveland Clinic, he said the health system has not deployed patient-facing or clinical AI without a human in the loop, underscoring the need to align new tools with how clinicians actually work.
8. Resistance isn’t always opposition — it’s often legitimate uncertainty. Leaders consistently framed resistance as a spectrum: personal discomfort with AI, ethical concerns and practical workflow realities.
Dr. Singh broke it down simply: Some people use AI personally, others don’t, while some have “very reasonable concerns” about harm to patients or broader impacts. His approach starts with transparency, listening and shared facts — plus making risks “measurable.” He described forums where staff can bring ideas and concerns, and a learning-health-system mindset: implement, measure, iterate and pivot — even away from AI if needed.
Mr. Odeh’s view is similar in practice: When solutions are deployed, the team invests time explaining how tools work, addressing concerns and gathering feedback — backed by a major emphasis on AI literacy.
9. AI literacy is not a “nice to have” — it’s the scaling strategy. If you want AI adoption without chaos, you can’t rely on a small central team. You have to raise the baseline across the organization.
Cedars-Sinai made democratized access a pillar, and Mr. Odeh said the system has trained over 1,000 staffers in AI literacy. As they scale, he described “deeper partnership with HR” and broader change management support so colleagues can “use AI safely and effectively while freeing time for higher-value work.”
10. Many AI requests aren’t AI at all — and the top AI leader becomes the chief problem-framer. A repeated first-year surprise: Once an organization gets excited, everything becomes “AI,” even when the best fix is workflow redesign, rules-based automation or better use of existing systems.
Mr. Chaudhary said that in his environment, “everyone wants AI,” but “75% of the time it’s not AI.” Often it’s broken processes, or “just optimizing a workflow in Epic” or a simpler algorithmic solution. He positioned the job as fundamentally cross-functional: “This role is … not technology. More of this is actually about relationships,” listening, learning the business problem and matching solutions appropriately.
Dr. Singh agreed that adding AI to a broken process won’t fix it — “technology can’t solve problems alone.”
11. Early wins matter — but the hardest part is organizing the path from pilot to “real.” Even where enthusiasm is high, leaders described a messy middle: procurement scrutiny, data protections, unclear pathways and governance that can feel slow.
Mr. Chaudhary described friction when requests arrive attached to a preferred external solution — then privacy, compliance and information security review is required. He said the system is working toward a more organized, methodical process, because today it can “cause confusion … and actually frustration.”
At the same time, tangible success stories change the internal conversation. Mr. Chaudhary pointed to ambient documentation as a turning point: more than 500 physicians using the tool, with clinicians telling him they can’t imagine going back because “pajama time is gone.”
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.