Balancing Innovation with Quality Care
Community health clinics (CHCs) are at a critical juncture. Entrusted with the care of some of the country’s most medically and socially complex populations, they are simultaneously navigating tightening budgets and evolving reimbursement models. In this landscape, agentic AI—a powerful combination of AIs and expert systems—is being heralded as a potential game-changer.
Yet, as Mathias Kolsch, Chief Scientist at Foresight, cautions, CHCs must tread carefully, distinguishing between autonomous systems and AI-assisted workflows that keep the human in charge. Indeed, any implementation of agentic AI must remain aligned with the tenets of value-based care (VBC).
From Prediction to Action: The Analytics Continuum in CHCs
CHCs already leverage predictive analytics to manage high-risk populations. These models synthesize clinical, claims, and other data sources to flag patients likely to face hospitalization or costly complications, enabling proactive alerts. Prescriptive analytics takes it a step further by recommending specific interventions—adjusting medications, scheduling home visits, or reallocating care-management resources.
Impactability modeling refines these efforts by identifying which at-risk patients are most likely to benefit from outreach. For example, Medicare case management programs have found that targeting impactable patients—rather than merely the highest-risk—yields greater reductions in readmissions and overall costs. In these scenarios, while AI provides the insight, human clinicians still make the final decisions and take action.
However, Agentic AI aspires to push beyond this analytic pipeline into true automation. In principle, an agentic system would sense real-time data, decide on actions according to its objectives, and then act automatically. For example, instead of generating an alert that a patient’s risk has increased, a truly agentic system might autonomously triage that patient, adjust care plans, and coordinate follow-up steps without manual prompts.
Understanding Agentic AI: Beyond Automation
Pioneering AI scientists Franklin and Graesser define an autonomous agent as “a system situated within and a part of an environment that senses that environment and acts on it, over time, in pursuit of its own agenda and so as to effect what it senses in the future.” Michael Wooldridge similarly describes agents as operating in a continual “sense–decide–act” loop, continuously perceiving the world, making decisions (often based on internal goals), and taking actions. In essence, an agent continually monitors its inputs, applies reasoning or a policy, and enacts outputs that change the state of the world. This cycle of autonomous perception and action – coupled with goal-driven behavior – is the heart of what distinguishes an agentic system from a fixed program.
Building the Infrastructure
Like all sophisticated AI, agentic systems require access to comprehensive, timely, and interoperable data—something many CHCs currently struggle with. Fragmented electronic health records, limited access to claims data, and underfunded IT infrastructure all pose major barriers. Without integrated clinical, claims, and social determinants of health (SDOH) data, even the most advanced models become ineffective.
This foundational requirement cannot be overstated: agentic AI can only succeed if it's built on solid data infrastructure.
A Multi-Layered AI Architecture for CHCs
Assuming data readiness, CHCs would need a layered, purpose-built technology stack, to begin its agentic journey. The goal isn’t technical sophistication for its own sake, but real, measurable improvements: reduced costs, enhanced care quality, improved patient engagement, and decreased disparities.
1. Foundational Data Architecture: Designed for interoperability and VBC outcomes, this layer ingests and harmonizes:
- EMR Data – Clinical insights from EHRs
- Claims Data – Utilization and cost information
- Quality Metrics – Aggregated performance indicators like HEDIS
Platforms like Caliper automate data ingestion and normalization, ensuring data consistency across sources.
2. AI Intelligence Core: The analytical engine powering agentic decision-making, incorporating:
- Cost Optimization Models
- Engagement Models
- Quality Improvement Models
- Disparity Reduction Models
These models guide intelligent workflows and learning loops.
3. Data Interaction Layer: Enables real-time, bi-directional data flow through:
- APIs, FHIR interfaces, HL7 listeners
- Event-driven subscriptions to EHRs, registries, SDOH feeds
This ensures the AI remains continuously informed and responsive.
4. Intelligent Agent Layer: Modular agents designed for closed-loop VBC tasks:
- Monitoring Agents – Track utilization, cost, and quality
- Engagement Agents – Trigger outreach and education
- Reconciliation Agents – Resolve discrepancies between payer and provider data
- Coordination Agents – Suggest next-best actions for care teams
These agents adapt over time, improving based on performance data.
5. Application Layer: SaaS Interface: A user-friendly that brings AI insights to life:
- Care Teams – Tools for patient engagement and gap closure
- Administrators – Dashboards for contract performance and reporting
- Payers – Visibility into population impact and provider effectiveness
Designed to ensure insights translate into meaningful, equity-focused actions.
Governance, Ethics, and Oversight: Guardrails for Agentic AI
Despite its capabilities, agentic AI is not a panacea. As a form of narrow AI, it requires robust oversight. CHCs must set clear boundaries for autonomous actions versus those requiring human review. Ethical considerations—data privacy, bias mitigation, and fairness—must be built into every stage of design and deployment.
With increasing federal focus on cost containment and operational efficiency, CHCs adopting agentic AI will face growing scrutiny around return on investment and demonstrable improvements in care quality.
Conclusion: Technology in Service of Equity and Impact
Ultimately, AI systems are only as good as the data they are built on. For CHCs serving diverse populations, ensuring that data is representative and algorithms are equitable is not just best practice—it is a moral imperative. We encourage CHCs to embrace the full continuum of analytics—from diagnostic to predictive to prescriptive—backed by actionable impactability modeling. Doing so often requires reconfiguring care teams and workflows so that the human-in-the-loop is strategically positioned to act on insights in real time. This integration enables a cycle of continuous learning and performance improvement between AI models and human action.
Unless agentic AI supports and enhances this process—optimizing for cost, quality, engagement, and disparity—it risks being little more than buzz without bite.
References:
- CIO:What makes a true AI agent?
- Franklin and Grasser: Is it an agent or just a program?
- Wooldridge, Michael Introduction to multiagent systems
- McKinsey: The Next Wave of Healthcare Innovation
- IBM: Prescriptive Analytics in Healthcare
- Impactability Modeling in Case Management
- HealthTech Magazine: What is Agentic AI?
Acknowledgement: What agentic AI actually is: a deeply researched and definitive explanation, Kane Simms
