It’s no secret that AI agents are the defining technology narrative for the year 2025. Earlier this year, we studied how AI agents could set the foundation to bridge the cold precision of data systems and the warm complexity of humanity to unlock new opportunities in healthcare.
The question isn’t whether AI agents will transform healthcare, it’s how quickly we can navigate the gap between promising pilots and production-scale reality. More than 83% of top execs from U.S. healthcare providers are already experimenting with generative AI solutions, yet the journey from proof-of-concept to meaningful impact remains fraught with challenges that are uniquely human in nature.
We think that AI agents can revolutionize healthcare by addressing critical challenges across the ecosystem: By streamlining data exchange between disparate systems, AI agents can enhance interoperability and enable analysis of comprehensive patient data, which enables unprecedented levels of care personalization. This could enhance value-based care (VBC) models that focus on driving better patient outcomes with improved care coordination across patients, providers, and payers.
The Interoperability Promise
At its core, an AI agent is a large language model (LLM) equipped with tools, operating within defined guardrails to achieve specific goals autonomously. In healthcare, this definition takes on deeper meaning when we consider what those tools might enable. Agents can serve as connective tissue between fragmented systems, surfacing insights from siloed data to create more personalized care experiences and better measure outcomes under value-based models.
But first, we must acknowledge a real challenge about healthcare data: it’s largely messy, scattered, and untapped. Interoperability challenges are data challenges. For example, different formats and standards are used to store healthcare data, with added complexity transitioning from existing standards (e.g. HL7 v2) to newer standards (e.g. HL7 FHIR). This is further exacerbated by the lack of alignment on data definitions and context (e.g. different descriptions recorded for the same procedure or prescription, like 100mcg daily vs. 50mcg x 2 tablets)
Modern healthcare data spans electronic health records (EHRs), revenue cycle management systems, and countless other provider and payer platforms, all speaking different languages. Even within these systems, only 20% of data exists in structured, easily shareable formats. The remaining 80% comprises emails, clinical notes, medical images, voice recordings that sit unstructured – rich with context but inaccessible to traditional integration approaches.
This is where AI agents’ multimodal capabilities become transformative. Unlike the structured APIs that dominate today’s data exchange paradigm, AI agents can process vast volumes of inputs in real time, opening up previously untapped sources of insights to act on them autonomously.
A Hybrid Future
That said, the future we’re building toward isn’t one where agents replace existing integration methods, but where they complement them. While agentic integration has the benefits of being multimodal, more adaptable, and faster compared to its structured counterpart, it remains non-deterministic in nature and could be costly to manage. In a regulated industry with sensitive data like healthcare, structured APIs will continue to be relied upon for precision and reliability. Nevertheless, AI agents’ multimodal capabilities can fill the gaps of structured APIs to surface insights from previously untouchable sources and focus on broader scopes of jobs to be done. The hybrid integration approach blending both APIs and agents can tap into the best of both worlds. We spoke with a founder who’s already working on this hybrid approach: an agent that pulls structured data from EHRs while simultaneously capturing voice inputs to support care coordination.
Anthropic’s MCP framework is making this vision even more tangible. At the very high level, MCPs are universal connectors – like the USB-C connections allowing agents to interface with any system, whether it’s a web service or a local file. Since its launch in late 2024, MCP has since taken the tech world by storm and is now emerging as a standard: many agentic platforms are now expected to host or even offer plug-and-play MCP servers. Taking it further, Google also released an open source standard, A2A, in April 2025 to enable agent-to-agent collaboration without sharing memory, resources, and tools.
These frameworks are still evolving rapidly and early in adoption, but they surface both the immense challenges and opportunities ahead: standardization questions, security concerns, scalability issues, and the promise of truly interoperable systems.
Looking ahead: AI-enabled Integrative Healthcare Experience
The agentic healthcare ecosystem taking shape spans both application and integration layers, touching B2B and B2B2C/B2C markets alike. As we observe this space, we’re excited about working towards an integrative, personalized experience that could fundamentally augment the ways millions of consumers currently manage their health. Several themes stand out to us as opportunities:
We see a world in which we no longer will rely on manual processes to make sense of the data separated between EHRs, pharmacies, wearables, claims, or labs. Now, insights into this data will be automated and delivered directly to the provider. Within value-based care models, that means bringing to light opportunities for spending and savings interventions, a path that has been paved by Aledade and will be followed by more AI-native models.
We are particularly excited about the potential for augmented AI processes to bridge the gap between clinical work and outcomes, particularly as AI patient-facing tools increase engagement outside of the doctor’s office. Double clicking into voice AI solutions for example, we observe that voice is increasingly becoming a critical component in shaping UI/UX. Players like Hippocratic AI are already deploying agents for chronic care management where patients can speak with agents verbally to monitor adherence and lifestyle adjustments. Newer players like Assort Health are building capabilities for specialized practices to engage with inbound patient calls, leaving more time for doctors to spend with their patients.
As multimodal capabilities expand, so will the need for specialized integrations. Companies like Care Quality and Health Gorilla create a national framework for data sharing with startups like Particle Health helping providers make sense of that data. But we see a world moving from just read capabilities to read/write capabilities. We’ve seen this world play out in finance, and we have no doubt it’s coming to healthcare.
Zooming out, competitive advantage in this agentic ecosystem will come from a few sources:
Data strategy represents the first frontier. Success will come from integrating siloed systems, leveraging diverse data sets, and achieving deep personalization at scale through improved memory and context retrieval capabilities. Core’s thesis: in a world of infinite data, context wins.
Agent architecture poses the second challenge. The future will involve many specialized agents working in concert, each designed for specific tasks but coordinated through careful orchestration. Robust architecture with guardrails, evaluations, and a human-in-the-loop (HITL) design will be critical. Core’s thesis: invest in AI + human coordination at scale
Go-to-market strategy will be the next distinguishing factor. Product-led growth models that demonstrate clear value early on, delivering ROI beyond efficiency gains (e.g. identify new revenue streams), will capture sustainable market positions. Following up with a land-and-expand strategy will enable deep integration within workflows that goes beyond just delivering point-solutions. Core’s thesis: embed in existing workflows, while limiting change.
Adaptability remains critical within a rapidly changing landscape. Foundational models, agentic infrastructure, and regulations are still evolving, while healthcare organizations remain inherently conservative in their adoption patterns. Building with flexibility at the core will be critical. Core’s thesis: don’t try to be the full-stack, foundational layers will outcompete.
Building Trust
Needless to say, the infrastructure shift is already underway, driven largely by horizontal platforms that define the technical foundations in standardizing agentic frameworks and communications protocol. Those who know how to build meaningfully on top of them will capture verticalized markets.
However despite technological advancement, real adoption barriers remain especially around trust. Healthcare is fundamentally about human relationships: between patient and provider, between institutions and the communities they serve. AI agents must earn their place to support these relationships instead of replacing them. There should be meaningful efforts dedicated to bridging any skills gap and addressing any disrupted or displaced labor.
Within healthcare, providing gradual exposure by starting with niche, high-value use cases in administrative workflows could help to lower the adoption barrier – building trust by starting small. After all, high accuracy is essential. The biggest customer pushback usually stems from risk aversion – the cost of error is high when human lives are at stake. Therefore clinical approval and HITL will remain important in the short term, beyond the baseline expectations of HIPAA and HiTrust compliance. This is a major reason why we backed Autoblocks to test and catch failures in AI tooling specifically for healthcare use cases.
Ultimately, there is a healthy tension in highly regulated spaces like healthcare – while AI can drive automation and hyperpersonalization, robust guardrails are essential to protect patient outcomes and data privacy. Understanding this balance of healthy skepticism and optimistic experimentation is key to realize technology’s potential in reshaping healthcare systems and everyday lives. The potential is enormous, now let’s go capture it.
