The Wild West of Healthcare AI Agents: Why We Need Industry-Specific Platforms

In my previous newsletters, I discussed how EMR interfaces will disappear and be replaced by conversational AI. While this vision resonated strongly with healthcare professionals worldwide, it has also inadvertently highlighted a growing problem: the chaos of uncontrolled AI agent proliferation in healthcare.

The Integration Nightmare in Healthcare IT

True end-to-end automation in healthcare remains virtually impossible with today's legacy systems. Look at any hospital's IT landscape: EMRs, PACS systems, laboratory information systems, pharmacy systems, scheduling software – they're all loosely connected through a patchwork of integrations that makes it nearly impossible to get a real-time view of patient care.

For decades, the healthcare IT world has been chasing the dream of seamless integration. We've built an alphabet soup of standards and protocols: HL7, FHIR, XDS, DICOM, APIs, web services, microservices – each promising to be the bridge between our islands of automation. But here's what nobody talks about: the "technical-legal integration paradox."

Let me explain. It's not just that these standards are too technical or low-level. The real problem runs deeper. Major EMR vendors like Epic might offer technical capabilities for integration, but their terms and conditions often restrict or prohibit actual integration attempts.

The "technical-legal integration paradox" in healthcare IT refers to the contradiction where electronic medical record (EMR) vendors, like Epic, provide technical capabilities for data integration but impose restrictive legal and contractual terms that hinder actual interoperability, creating a gap between the promise of seamless integration and the reality of constrained data sharing due to business incentives and regulatory complexities.

It's a fascinating paradox that reminds me of my time at SAP back in 2006. During my SAP days, we had extensive discussions about developing new APIs (called BAPIs in SAP terms). While we had highly functional APIs, I witnessed firsthand how the lead product management team at headquarters resisted prioritizing certain API developments. Their concern? That easier integration might lead to fewer licensed users, threatening the revenue stream. It was a valid business concern, but it reflected a protective, closed mindset that ultimately hurt innovation.

Today, we're seeing the same pattern repeat in healthcare IT. The result? Organizations end up building redundant systems that replicate business functions, forcing us to constantly translate between fundamentally different languages. An EMR's understanding of a patient visit becomes something entirely different in a billing system or a pharmacy system. It's like having three different people describe the same event in three different languages, something always gets lost in translation.

The result? Brittle integrations that only automate a portion of the workflow, constant maintenance headaches, and systems that break whenever there's an update. We've essentially created software-defined departments instead of truly integrated healthcare organizations.

The Current State of AI Agents in Healthcare

Let's look at what's already happening in hospitals today. We're seeing an explosion of specialized AI agents entering healthcare organizations:

• Clinical Documentation Agents, promising to eliminate manual note-taking
• Diagnostic Support Agents such as PathChat analyzing pathology images
• Radiology Assistants from multiple vendors providing image analysis
• Patient Communication Agents handling appointment scheduling and follow-ups
• Clinical Trial Matching Agents scanning patient records for trial eligibility
• Medical Literature Agents summarizing research papers and guidelines
• Laboratory Analysis Agents interpreting test results
• Drug Interaction Agents checking medication compatibility

Each of these agents is impressive in isolation. But here's the problem: they're entering healthcare organizations through various channels, often bypassing traditional IT governance. Some are being integrated directly into frontends, others are accessing data through shadow IT practices, and many are operating without proper security monitoring or data access controls.

The Promise of Agentic Platforms in Healthcare

But here's where it gets interesting. While these integration challenges might seem insurmountable, agentic AI platforms offer a fundamentally different approach. Instead of trying to force different systems to speak the same language, we can create a layer of AI agents that understand and translate between these different contexts naturally.

As Eric Topol eloquently pointed out in the Lancet a few days ago : 

"By harnessing these advances, AI agents have the potential to become valuable teammates to human clinicians... Instead of juggling multiple tools, the clinician could interact with a single manager agent." 

This vision perfectly aligns with what we're building - a future where AI clinical manager agents orchestrate multiple specialized tools while maintaining a holistic understanding of patient care.

Imagine a healthcare environment where multiple agents work together, each with specific authority and access rights, supervised by healthcare professionals, and guided by healthcare-specific metrics like patient outcomes, quality of care, and operational efficiency. These agents won't just automate individual tasks, they'll work in coalition with each other, responding to changes in patient conditions, adapting to new clinical guidelines, and optimizing for both clinical and operational goals.

While I've previously argued that openEHR and AI agents will replace traditional EMRs, and I stand firmly behind this prediction, we need to be pragmatic about the transition period. Healthcare organizations can't switch off their legacy systems overnight. Instead, we'll see a gradual transformation where agentic platforms first connect to existing systems, then progressively take over their functions, and ultimately replace them with more intelligent, flexible architectures built on openEHR principles.

The key difference from previous integration attempts is that these agents can learn. They'll learn from existing systems, from healthcare professionals' actions, and from each other. They can capture the "tribal knowledge" that makes healthcare work, all those unwritten rules and practices that could never be effectively coded into traditional systems. As this learning progresses, it will accelerate the transition away from legacy EMRs toward more intelligent, agent-based systems.

The hidden dangers

First, there's the security nightmare. IT departments are losing visibility into what data these agents are accessing, how they're processing it, and where it might be aggregated. When an AI agent is integrated directly into a department's workflow without proper oversight, who's monitoring the data flow?

Second, we're creating new data silos. Each agent operates in its own bubble, using its own data models and interfaces. One agent's understanding of a patient's condition might completely differ from another's, leading to potentially dangerous inconsistencies.

Third, there's the compliance challenge. How do you maintain HIPAA compliance when you can't track how patient data is being used across multiple independent AI agents? How do you ensure GDPR's right to be forgotten when you don't even know where all the data resides?

Why Healthcare Needs Industry-Specific Platforms

This is precisely why we need healthcare-specific agentic platforms. The requirements for healthcare are fundamentally different from other industries. We're not just handling business data – we're dealing with patient lives.

A proper healthcare agentic platform needs to provide:

1. Centralized Security and Governance: Every AI agent interaction needs to be monitored, logged, and controlled. IT departments need visibility into data access patterns and the ability to enforce security policies consistently
2. Clinical Context Preservation: Unlike general-purpose AI platforms, healthcare platforms must understand and maintain clinical context across all agent interactions. A lab result isn't just a number – it's part of a patient's clinical narrative.
3. Regulatory Compliance: Healthcare-specific compliance requirements need to be built into the platform's DNA, not bolted on as an afterthought.

The Platform Approach

At Isaree, we believe the solution lies in creating an open platform specifically designed for healthcare AI agents. Rather than fighting the proliferation of agents, we need to channel it through proper infrastructure.

Think of it like a city's infrastructure. Instead of letting every company dig up roads to lay their own pipes and cables, we create an open shared infrastructure that everyone can use safely and efficiently. Our platform provides the secure foundations that all healthcare AI agents can build upon, while ensuring proper governance, security, and interoperability.

This approach allows healthcare organizations to:

• Maintain central control over data access and security
• Monitor and audit all AI agent interactions
• Ensure consistent data governance
• Maintain regulatory compliance
• Enable safe interoperability between agents

A Call to Action

The proliferation of AI agents in healthcare is inevitable and, in many ways, desirable. But we need to get the infrastructure right. Healthcare organizations need to start thinking about AI governance now, before the situation becomes unmanageable.

We're already seeing the consequences of uncontrolled AI agent adoption in some hospitals, shadow IT practices, security vulnerabilities, and data governance nightmares. The time to act is now.

We want to understand you better and have prepared a survey for you to participate in. 

*Bart de Witte is the founder of Isaree, a Berlin-based company building the next generation of AI-powered healthcare platforms and the founder of the Hippo AI Foundation, a non profit focusing on advancing open source AI in medicine . For more insights into the future of healthcare IT, follow him on LinkedIn or subscribe to this newsletter.*