How Agentic AI Could Transform Healthcare’s Broken Workflows

According to MIT Technology Review, current ambient AI assistants that gained mainstream traction in 2023 can record, structure, and summarize patient encounters in real time, with Nabla’s system achieving accuracy in the “high 90s” percentage range. Dr. Lee reports that what previously took 45 minutes of documentation now happens instantly, allowing full patient attention and better eye contact. Nabla CEO Alexandre LeBrun notes patients respond positively to the technology, appreciating their physician’s undivided attention. The company is now developing agentic AI that will link multiple clinical tools into unified workflows, with capabilities including automated pre-charting, EHR integration, and medical coding. This evolution represents a significant step beyond current AI assistants toward comprehensive clinical workflow automation.

The Silent Crisis in Clinical Workflows

What the source article touches on but doesn’t fully articulate is the magnitude of healthcare’s workflow crisis. The average physician spends approximately two hours on EHR work for every hour of patient care, creating what researchers call the “third shift” phenomenon where clinicians complete documentation after hours. This isn’t just an inconvenience—it directly contributes to physician burnout, which affects nearly 50% of clinicians according to recent studies. The promise of agentic AI isn’t merely about saving time; it’s about addressing a systemic failure in how healthcare technology has evolved, where each new system added complexity rather than reducing cognitive load.

The Technical Hurdles Ahead

While the vision of unified agentic systems is compelling, the technical implementation presents substantial challenges. Healthcare systems operate on decades of legacy infrastructure that wasn’t designed for interoperability. An AI agent that can seamlessly navigate between different EHR systems, billing platforms, and clinical decision support tools requires solving integration problems that have plagued healthcare IT for years. The “composable agents” concept faces real-world constraints around data standardization, privacy regulations, and institutional resistance to changing established workflows. Most healthcare organizations have between 15-20 different clinical systems that rarely communicate effectively, creating the exact problem agentic AI aims to solve.

The Trust Imperative in Clinical AI

LeBrun’s emphasis on trust touches on the fundamental barrier to AI adoption in healthcare. Unlike other industries where AI errors might mean recommending the wrong movie, in healthcare, mistakes can have life-or-death consequences. The “high 90s” accuracy rate mentioned, while impressive, still leaves room for potentially dangerous errors in complex medical contexts. This explains why companies like Nabla emphasize conservative approaches and human oversight. The real test for agentic AI will come when these systems handle more complex clinical decision support, where the line between assistance and autonomous decision-making becomes dangerously blurry. The medical community’s cautious approach reflects lessons learned from earlier AI implementations that promised more than they delivered.

The Implementation Reality Check

The transition from ambient AI to true agentic systems represents a quantum leap in complexity. Current AI assistants primarily handle documentation—a relatively straightforward task compared to the multi-step, context-dependent workflows that agentic AI promises to manage. The cardiology example LeBrun provides sounds revolutionary, but it assumes perfect real-time integration across systems that often have different data formats, update schedules, and access protocols. Early implementations will likely face significant limitations, requiring extensive customization for different medical specialties and practice settings. The gap between demonstration projects and widespread clinical utility has historically been substantial in healthcare technology adoption.

The Competitive and Regulatory Landscape

What the source doesn’t mention is the emerging competitive battlefield in clinical AI. While Nabla appears focused on the agentic approach, major EHR vendors like Epic and Cerner are developing their own AI capabilities, potentially creating walled gardens that resist third-party integration. Meanwhile, regulatory bodies like the FDA are still determining how to classify and oversee AI systems that demonstrate true agency in clinical workflows. The companies that succeed will need to navigate not just technical challenges but complex regulatory requirements and established vendor relationships that dominate healthcare IT purchasing decisions. The path to widespread adoption will likely be slower and more fragmented than the technology’s potential suggests.

The Human Factor in AI Transformation

Perhaps the most insightful aspect of the source material is Dr. Lee’s emphasis on education and cultural change. Technology implementations in healthcare often fail not because of technical limitations, but because they don’t account for human factors. Clinicians who’ve spent decades developing their workflow patterns may resist systems that fundamentally change how they practice medicine. The success of agentic AI will depend as much on change management and clinician buy-in as on technical capabilities. Organizations that treat this as purely a technology implementation rather than a cultural transformation will likely see limited adoption, regardless of how impressive the underlying technology might be.