Healthcare data leaders are warning that outdated technology platforms are slowing the adoption of artificial intelligence in hospitals and health systems across the United States. As AI applications evolve from predictive analytics to more autonomous “agentic” systems capable of initiating actions without direct human input, experts say the foundation beneath these tools must be modernized to avoid bottlenecks, security risks and ethical oversights.
The caution comes amid growing investment in AI for clinical decision support, operational efficiency, and patient engagement. Yet many institutions still rely on legacy electronic health record (EHR) systems and fragmented data warehouses that were not designed to handle the real-time, high-volume data flows required by advanced AI models. Without upgrading these core infrastructures, even the most sophisticated algorithms may fail to deliver reliable results at scale.
Chuck Podesta, Chief Information Officer at Renown Health in Reno, Nevada, emphasized during a recent industry panel that legacy platforms often lack the interoperability and processing speed needed for agentic AI — systems that don’t just analyze data but can trigger workflows, such as automatically scheduling follow-up tests or adjusting medication orders based on patient vitals.
“If your data foundation is brittle or siloed, you’re not just slowing innovation — you’re introducing risk,” Podesta said. “Agentic AI requires trustworthy, timely data. If the platform can’t deliver that consistently, clinicians won’t trust the output, and patients could be harmed.”
Roshan Hussain, Senior Vice President and Chief Data Analytics Officer at RWJBarnabas Health in New Jersey, echoed this concern, noting that many health systems are attempting to layer AI tools onto outdated architectures without addressing underlying data quality or governance gaps.
“You can’t bolt a Ferrari engine onto a Model T chassis and expect it to perform,” Hussain explained. “We’re seeing organizations invest heavily in AI models while neglecting the data pipelines that feed them. That leads to biased outputs, delayed insights, and compliance challenges — especially when AI starts making autonomous decisions.”
Sarang Deshpande, Vice President of Data and Analytics at Franciscan Alliance in Indiana, added that the pressure to modernize is intensifying as Epic Systems, the dominant EHR vendor in the U.S., extends its software update cycles. Longer build cycles signify health systems wait longer for new features and integrations, prompting some to consider in-house development or alternative platforms to maintain agility.
“When your core EHR updates every 18 to 24 months instead of annually, you lose the ability to rapidly test and deploy AI use cases,” Deshpande said. “That’s driving interest in cloud-native data platforms like Databricks and Snowflake, which offer greater flexibility for integrating diverse data sources and supporting machine learning workflows.”
The shift toward cloud-based data lakehouses — architectures that combine the scalability of data lakes with the reliability of data warehouses — has gained traction in healthcare over the past three years. These platforms allow organizations to store structured and unstructured data (such as clinical notes, imaging metadata, and wearable device streams) in a single environment, enabling more comprehensive AI training and validation.
Databricks, founded by the creators of Apache Spark, has partnered with several major health systems to deploy its Lakehouse Platform for use cases including sepsis prediction, readmission risk modeling, and supply chain optimization. Snowflake, meanwhile, has highlighted its role in enabling secure data sharing between providers, researchers, and public health agencies through its Healthcare and Life Sciences Data Cloud.
Yet, experts caution that migration to these platforms is not a plug-and-play solution. Successful implementation requires careful planning around data governance, staff training, and change management. Without clear policies on data access, model accountability, and bias mitigation, even modernized infrastructures can introduce new risks.
To address this, some health systems are adopting “fast-track” AI governance models designed to accelerate innovation while maintaining ethical safeguards. These frameworks often include automated monitoring for model drift, predefined escalation paths for high-risk AI decisions, and regular audits by multidisciplinary ethics committees.
“Governance shouldn’t be a bottleneck,” Hussain noted. “But it similarly can’t be an afterthought. The best models embed checks and balances into the development lifecycle — from data sourcing to deployment — so that speed and safety go hand in hand.”
The push for modernization is also being shaped by federal initiatives. In 2023, the Office of the National Coordinator for Health Information Technology (ONC) released its final rule on interoperability and information blocking, which includes provisions aimed at improving access to electronic health information for AI development and public health reporting.
Under the rule, health IT developers must standardize APIs to allow easier data exchange, a requirement that supporters say could reduce friction in building AI-ready data pipelines. The ONC estimates that compliant systems could save the U.S. Healthcare system up to $30 billion annually through reduced administrative burden and improved care coordination — though actual savings depend on widespread adoption and effective implementation.
Meanwhile, the Food and Drug Administration (FDA) has increased its oversight of AI/ML-based software as a medical device (SaMD), issuing draft guidance in 2024 on predetermined change control plans — a mechanism allowing manufacturers to pre-specify planned modifications to AI algorithms without requiring a new premarket submission for each change.
This approach aims to balance innovation with patient safety by enabling iterative improvements to AI models while maintaining regulatory oversight. The FDA has received over 200 submissions referencing such plans since the guidance was issued, reflecting growing use of adaptive AI in radiology, cardiology, and pathology.
Despite these advances, challenges remain. A 2024 survey by the American Hospital Association found that only 38% of hospitals feel “very prepared” to integrate AI into clinical workflows, citing data quality, workforce readiness, and unclear liability frameworks as top concerns.
“We’re at an inflection point,” Podesta said. “The technology is ready. The use cases are clear. But if we don’t fix the foundation — the data platforms, the governance, the trust — we’ll keep seeing pilot projects that never scale, or worse, AI systems that make decisions based on flawed or incomplete information.”
For healthcare leaders, the message is clear: investing in AI without modernizing the underlying data infrastructure is like building a house on sand. The next wave of innovation won’t just depend on better algorithms — it will hinge on whether health systems can create resilient, secure, and agile data ecosystems capable of supporting the demands of autonomous intelligent systems.
As organizations evaluate their next steps, many are turning to peer networks and industry consortia for guidance. Groups like the Healthcare Information and Management Systems Society (HIMSS) and the College of Healthcare Information Management Executives (CHIME) regularly publish best practices on data modernization and AI readiness, offering frameworks that balance technical rigor with clinical utility.
The next major checkpoint in this evolving landscape is the ONC’s scheduled release of updated interoperability benchmarks in early 2025, which will measure progress toward nationwide health information exchange. Health systems aiming to stay ahead in the AI race will be watching closely to see how these metrics influence vendor roadmaps and federal incentives.
What steps is your organization taking to prepare its data infrastructure for the era of agentic AI? Share your insights in the comments below, and assist spread the conversation by sharing this article with colleagues in healthcare IT, clinical leadership, and public health.