AI agents: a new era in healthcare. Or are they?

They’re already on the phones. In the charts. Behind the front desk. Reading your vitals and drafting your care plans.

A new era in healthcare technology? Seems so.

It’s the perfect storm

Physicians are burnt out.

And not just tired. They are systemically maxed out because of perennial staff shortages and the pressure to do more with less to save costs. 

Precisely for that reason, the AI agent technology (and AI overall, for that matter) is picking up steam:

• With $59.6 billion invested, AI-led venture funding in Q1, 2025.
• Domain-specific LLMs are becoming smarter, while the progress in multi-agent systems can now power AI agents to collaborate across tasks.
• Ambient AI tools like Suki AI, Abridge AI, and Commure are catching fire in healthcare, powering AI agents with situational awareness.
• The number of physicians whose enthusiasm beat their concerns about healthcare AI increased to 35% in 2024 from 30% in 2023.

The stars are truly aligned for AI agents.

From the model of scarcity to the model of abundance

For years, AI has been waiting for direction. Now geared up with advancements in reasoning and autonomous thinking, AI agents can scale healthcare resources that have not existed in the past.

🏥 Administrative automation

Among all use cases, administrative automation is one of the areas with the highest potential for AI automation. That’s why startups in the administrative AI space account for 60% of total AI investment in healthcare since 2021. 

For AI agents, the potential applications are extensive, too:

• Innovaccer’s voice-activated AI agents interact with patients for scheduling, protocol intake, referrals, prior authorization, care gap closure, and HCC coding.
• Focused on revenue cycle management, Thoughtful.AI’s AI agents CAM, EVA, DAN, and the rest do the heavy lifting of eligibility checks, claims processing, payment posting, and other related tasks.
• VoiceCare AI launched a pilot with the Mayo Clinic to automate provider-care conversations.

🥼 Care coordination and clinical decision-making

Timely triage and effective care coordination are a tough bar to scale, given the sheer number of patients. With AI agents, healthcare providers can put their best care forward across a large patient base:

• One of Sully.AI’s agents, Sully the Nurse, helps nurses scale their capabilities with automated symptom triage.
• In partnership with Hippocratic AI, WellSpan debuted a clinical GenAI healthcare agent that has already reached over 100 of WellSpan’s multilingual patients to improve access to life-saving cancer screenings.
• KATE AI, a Mednition startup, aids emergency nurses with ED triage, identifying sepsis and non-sepsis patients with an exceptional accuracy of 99%.

🧑🏻 Patient engagement

Poor patient engagement directly translates into poor health outcomes and higher healthcare costs. However, ensuring active listening across diverse populations and healthcare settings is an arduous task — yet, it’s a task AI agents can handle.

• Cedar, a patient financial platform for healthcare providers, unveiled Kora, an AI voice agent that handles patient billing calls.
• Hello Patient is integrating generative AI phone agents with EHR systems to automate patient-facing communication work.

💊 Drug discovery and clinical research

The complexity and costs of clinical research trials have reached an all-time high. With a clear need to manage the complexity, life sciences companies have also gotten in on the AI act.

• Grove AI’s agent, Grace, expedites the enrollment of patients in trials by calling patients as soon as they express interest in a trial.
• Manas AI cuts the timeline and costs of drug development by identifying high-potential therapeutic candidates.
• Backed by a whopping $1 billion, Xaira Therapeutics is readying AI-generated drugs. 

Where are AI agents headed?

AI agents are beginning to draft their own to-do lists in healthcare. But it’s still a far cry from what these systems may soon be capable of.

Agentic and multi-agent systems

Most AI agents are semi-agentic at best. Either task-oriented or prompt-based, they lack proactivity and can’t implement multi-step strategies on their own.

In an agentic framework, a central large language model (LLM) orchestrates specialized sub-agents (other AI models or tools) to nail complex, multi-step tasks with more accuracy.

• Hippocratic AI’s constellation architecture revolves around a primary LLM paired with 20+ task-specific support models for cross-checking.
• Google rolled out a centralized hub where companies can tap into pre-built AI agents and a communication protocol for agent-to-agent collaboration.

In the same vein, findings suggest that multi-agent systems — having multiple AI agents collaborating autonomously — beat single-agent systems at mortality prediction accuracy and transparency.

Specialized LLMs and VLMs

Healthcare AI agents rely on domain-specific large language models, either proprietary or fine-tuned, to achieve accuracy and clinical reliability. 

However, healthcare isn’t just about text. Clinical photos, X-rays, and MRIs should be integrated alongside text for a complete clinical image.

VLMs, or voice language models, enable multimodal understanding of AI agents, allowing them to respond to a mix of text, visuals, and structured EHR data.

• MONAI Multimodal, an open-source framework for medical imaging, leverages multimodal AI models to support cross-modal data integration.

EHR foundation models

Once given access to EHR data, AI agents obtain the rich clinical context they need to reason accurately about a patient’s health journey. That’s why AI agents are being integrated into EHR platforms, whether through foundation models or directly, to both fetch data and automate documentation.

• Oracle Cerner’s new Clinical Digital Assistant provides a voice-driven interface to EHRs.

A tiny dose of reality

Investors and healthcare players might be bullish on AI agents, but the path to mainstream autonomous intelligence is a long and bumpy one.

• The FDA's traditional medical device regulation was not intended for adaptive AI and ML. It means that even minor updates to AI and ML-driven devices may need a premarket review.
• Hospitals and vendors must secure data pipelines and often rely on federated learning or synthetic data to mitigate risk.
• AI agents can come across as smart by getting the final answer right. But they don’t yet have the medical know-how to back it up, which makes their decisions hard to trust without a human in the loop.
• Physicians say they need a reliable feedback loop, data privacy assurances, smooth EHR integration, increased regulatory oversight, and adequate training and education before they can commit to AI. 
• Once deployed, AI systems can degrade over time as clinical guidelines or patient populations evolve. Continuous pre-training can keep the healthcare knowledge current, but can only be implemented with extreme caution.

Thanks to AI, the doctor will see you now

AI agents aren’t just knocking at healthcare’s front door. They’re already there, automating clinical note-taking, handling patient outreach, and scaling the kind of attention clinicians scramble to give.

It’s not sci-fi just yet. Between regulatory hurdles, lack of trust, and data silos, the autopilot mode in healthcare is still a far-fetched concept. But the momentum of AI agents is undeniable.

Need a hand with LLM integration, continuous pre-training pipelines, prompt engineering, and everything AI- and data-related in between? Contact us to start and scale your AI journey.

Related reading:

• AI agents for healthcare: benefits and use cases (Oracle)
• AI agents are hospitals' newest 'employees.' We called their references (Newsweek) 
• Newsweek connected with 10 agentic AI developers to learn how the tech is changing health care (Fierce Healthcare)