Cut Through The Hype: 3 Things To Know Before Adopting Agentic AI

The healthcare industry has entered the “agentic AI era,” marked by a rapid increase in new AI startups and significant funding. Yet 80% of health systems say they lack the resources to identify, select and implement AI solutions.

So, here’s the rub: healthcare systems that could benefit most from agentic AI are often very slow or even unable to adopt it.

For over a decade, I’ve seen multiple “eras” of health IT evolution. “Agentic AI” is the next solution to help fix what healthcare has been focused on for years – connecting patients to care easily and efficiently, while reducing staff burden.

Yet, what is new and different is AI’s profound, and currently untapped, potential to completely transform how patients interact with healthcare, including on the administrative and communications side of patient access – scheduling, rescheduling, referrals, prescriptions, post-discharge, general questions and more.

Here are three considerations for healthcare organizations when adopting an agentic AI solution for patient communications:

Don’t be fooled by “integration-lite” capabilities

Every vendor will promise “deep” EHR integration because it’s the foundation of a truly valuable AI agent. The ability to securely read from and write to patient records is how an agent automates workflows and delivers real value.

But how do you know if a vendor is a true leader in this space, not just “getting by”? A key indicator is their adoption of new, forward-looking standards.

For example, consider the Model Context Protocol (MCP). This is a new, open standard developed by Anthropic that is fundamentally changing how AI agents interact with external systems. Think of it as an “API for AI agents” that fills the gaps where traditional APIs fall short for LLMs, enabling more secure, dynamic, and effective interactions with systems like EHRs.

Are your potential partners actively adopting MCP? If so, are they building a simple wrapper, or thinking through the end-to-end task and designing the MCP function to limit judgement calls by the LLM? Assess their team and investments: do they have the technical foresight and resources to do more than simply get by? The right partner will not just meet today’s integration needs but will position your health system to take full advantage of future advancements.

Avoid the quick fix trap

When adopting a new solution, leaders want a sustainable, long-term fix — not a temporary one. Yet, many new implementations fail to deliver lasting value because they only address the surface-level problem, leaving the core, infrastructure-level problems and manual work unsolved.

I’ve recently heard of agentic AI implementations that still rely on file transfers or manual data entry. Although this approach results in rapid deployment, it is a quick fix in the worst way. It perpetuates manual processes, increases the risk of the use of outdated data and/or risk of PHI/PII mishandling, and fails to free up administrative staff long-term.

In contrast, a true autonomous, agentic AI solution either uses dynamic integrations into downstream systems (e.g. via MCP) or naturally integrates into a broad portfolio of solutions already offered by the vendor. These approaches deliver a sustained return on investment and may involve slightly longer initial implementation.

Also, resist the urge to over-engineer agent prompts for a quick fix as some vendors may throw everything into an agent prompt in service of quick implementation. While this might seem efficient for a fast go-live, it’s brittle and introduces risk. Whereas thoughtfully designed intent-based MCP tools can increase performance, reduce the risk of hallucination and improve scalability.

There are also questions that should be answered to help provider organizations look ahead: will this implementation still be valuable in six months or a year? Is a focus on speed today sacrificing the deeper, more transformative value your system deserves tomorrow? These are important considerations as health providers weigh the value of speed for sustainability and safety.

Avoid the security “check box” mentality

There is no “finish line” for security and privacy, especially in the era of agentic AI; the landscape is in constant flux. When I first started in this industry, I believed HITRUST certification was enough. Today, I know it’s not. Such certifications are a snapshot in time, not a reflection of an ongoing commitment to protecting your health system’s most sensitive data.

While many vendors take the right steps — earning certifications, hiring security leaders, and implementing standard protocols — that should be the starting point. Healthcare data security and privacy have evolved far beyond what it was even a few years ago, and now agentic AI systems – which are constantly learning, adapting and making decisions on their own – compound this. As agentic AI introduces new security challenges, health system leaders should prioritize partners who not only have solid security measures in place today but are also actively committed to staying ahead of AI security trends and can rapidly adapt to new threats with effective solutions, such as mitigating data spillage using MCP or implementing test agents to analyze and score conversations.

Today, a firm commitment to security and privacy must be foundational and cultural. A security-first mindset and commitment must span technologies, all departments, and all people company-wide. It’s an organizational value, not a handful of certifications managed by a small team charged with “governing” the rest of the employees.

For instance, companies working with government agencies may pursue FedRAMP High authorization, the U.S. government’s most rigorous security standard. This is a smart move because it’s more than simply checking a box. The process itself further embeds a culture of security across the entire organization.

In closing, in the new agentic AI era, you aren’t just buying a piece of technology; you are adopting a system that will learn, adapt, and become a part of your team’s operations. An agentic AI vendor must be a partner that’s as invested in your long-term success as you are.

How do you tell the difference? A vendor sells you a solution and hands you a manual. A partner works with you to understand your specific workflows, co-creates a robust implementation plan, and provides ongoing support that goes beyond a standard help desk.

Ultimately, the most valuable AI agent is the one that’s backed by a partner who’s in it for the long run, not just the quick sale.

Photo: Yuichiro Chino, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.