Preparing for the Next Phase of AI Adoption in Healthcare
Trends * December 29, 2025
The next wave of AI in healthcare isn't about bigger models or flashy demos. It's about collaboration: getting large language models to work with trusted algorithms, data services, and clinical systems in a predictable way. The Model Context Protocol (MCP) is the connector that makes this practical.
MCP defines how LLMs, agent-based apps, and enterprise platforms plug into verified knowledge sources. That means cleaner integration, fewer one-off builds, and a faster path to voice-driven workflows that actually fit clinical practice.
Why MCP matters now
Many health systems still run on monolithic tech that slows AI adoption. MCP flips that script by letting teams pair EHR platforms with focused AI agents through defined APIs. You keep your core workflows and add targeted capabilities without deep, risky rewires.
Voice is the obvious early win. Ambient listening and voice commands can document visits, summarize encounters, and surface relevant data-then hand results back to enterprise systems through MCP. Less custom code. More interoperability. More control.
- Adopt best-of-breed tools while keeping EHR workflows intact
- Standardize how AI agents access curated knowledge and services
- Reduce integration effort and shorten deployment timelines
For an overview of the protocol, see the Model Context Protocol announcement from Anthropic: What is MCP?
Documentation quality moves center stage
Medicare audits are intensifying, especially in Medicare Advantage. The focus: does the chart support what's coded, and is the diagnosis being managed? Expect payers and providers to invest in validation tools that confirm clinical evidence in real time and at scale.
This isn't just compliance. Clean, complete documentation drives accurate reimbursement and better outcomes. The latest CMS Hierarchical Condition Category (HCC) model raises the bar for specificity and supporting evidence, which puts pressure on encounter documentation and coding workflows.
Learn more about CMS risk adjustment here: CMS Part C Risk Adjustment
- Deploy ambient tools to capture detail, then validate what's documented
- Use AI to flag missing evidence, unsupported diagnoses, and coding gaps
- Run pre-submission checks so data sent to Medicare is accurate and defensible
- Tighten feedback loops between coding, CDI, and clinical teams
As the transition to the newer HCC model completes in 2026, organizations that surface gaps at the point of care-and confirm clinical accuracy-will protect revenue and reduce downstream risk.
Smaller, domain-specific models will lead
Enterprises love the potential of large language models-until usage scales and costs spike. Token consumption across thousands of users adds up fast. Security reviews and GPU dependencies don't help either.
The practical move: smaller models tuned for clinical tasks. They run on standard CPUs, can be deployed inside your environment, and integrate through MCP so they work alongside existing systems. You get predictable costs, tighter data control, and reliable performance.
- Pair small models with curated knowledge sources via MCP and retrieval
- Prioritize use cases with clear ROI: documentation, coding support, orders, summaries
- Measure total cost of ownership across infra, licenses, and support
- Instrument models for accuracy, bias checks, and audit trails
What to do in Q1-Q2 2026
- Map top 3 use cases where voice, summaries, or validation would save hours weekly
- Ask vendors for MCP support and documented APIs; avoid closed integrations
- Stand up a documentation validation pilot tied to HCC specificity and audit readiness
- Trial a small model on CPUs inside your environment; compare accuracy and cost to your LLM baseline
- Set clear governance: human-in-the-loop, evidence requirements, and model monitoring
The bottom line
Healthcare is moving from "big, singular AI" to "modular, collaborative AI." MCP provides the connective tissue. Validation tools protect revenue and care quality. Smaller, domain-specific models make scale feasible and secure.
Build for interoperability, verify what goes in the chart, and keep costs in check. That's how AI becomes useful at the point of care-and sustainable for the enterprise.
If your team is leveling up AI skills for these initiatives, explore practical courses by role here: AI courses by job
Your membership also unlocks:
