Clinical Decision Support with AI Agents
Clinical Decision Support has always been about one central goal: helping clinicians make better decisions at the point of care. As AI agents begin to mature within healthcare, the workflow layer becomes especially critical. It is at this layer that AI transitions from an isolated tool to an integrated part of clinical practice.
The question is not whether AI can contribute to CDS. The question is how the agent-based model reshapes the very structure of CDS itself.
The Classic Model of CDS: Push and Pull
Historically, CDS has been designed around two modes of interaction: push and pull.
Push CDS refers to systems that proactively surface alerts or recommendations while the clinician is charting, ordering, or reviewing information. Common examples include drug–drug interaction alerts, allergy warnings, and rule-based checks embedded directly into the EHR. These mechanisms have been invaluable for patient safety. At the same time, they have contributed to what clinicians know all too well as alert fatigue.
Pull CDS, on the other hand, is initiated by the clinician. It involves searching for evidence, reviewing order sets, consulting guidelines, or exploring relevant literature. In this model, the clinician decides when to seek support, and the system responds.
For years, EHR design has revolved around this push–pull logic. One mode is interruptive and automatic. The other is intentional and on demand.
How AI Agents Change the Equation
If your understanding of AI in healthcare is shaped primarily by tools like ChatGPT, it is easy to assume that AI simply enhances pull-based CDS. You ask a question. The system answers.
But that is only a fraction of what is possible.
When viewed through an agent-based architecture, AI is not just a reactive question-answering system. An agent has access to data, tools, reasoning logic, and guardrails. It can operate proactively or reactively. It can monitor context, retrieve evidence, synthesize patient history, and execute actions depending on the workflow.
This is where the concept of the clinical copilot emerges.
The Clinical Copilot
A clinical copilot is not just a chatbot with medical knowledge layered on top. It is not a static Q&A engine.
It is one or more coordinated agents working together to support clinical decision-making in real time. It is a layered system capable of both pushing relevant insights into the workflow and responding intelligently when clinicians seek clarification.
More importantly, it can make those insights actionable.
Because agents can connect to tools and structured systems, a copilot can:
Order labs
Generate documentation
Summarize patient notes
Verify data consistency
Surface relevant clinical guidelines
Prepare prior authorization documentation
This shifts CDS from isolated alerts or search queries to a continuous, conversational experience embedded within the clinical workflow.
Rather than separating push and pull, copilots merge them into a single adaptive interaction model. The system listens, reasons, and contributes in context. It understands workflow state, applies safety guardrails, and operates within defined boundaries.
From Alerts to Conversations
Traditional CDS has been episodic. An alert fires. A guideline is consulted. A search is performed.
Agent-powered copilots create a more fluid model. They can observe patterns across encounters, track evolving patient trajectories, and surface insights at moments of high relevance without overwhelming the clinician.
The goal is not to replace physician judgment. It is to reduce cognitive load, synthesize complexity, and ensure that critical information does not remain buried within the EHR.
As agent-based architectures mature, supported by emerging standards such as Model Context Protocol and evolving conversational UI paradigms, clinical copilots will become increasingly central to care delivery. They will not function as standalone apps, but as embedded intelligence within existing workflows.
Rethinking CDS for the Agent Era
The transition from rule-based alerts to agent-driven copilots represents more than a technical upgrade. It is a conceptual shift.
CDS is no longer just about triggering reminders or enabling search. It becomes an ongoing collaboration between the clinician and the intelligent system. Push and pull dissolve into a unified, conversational layer of support.
That is the direction we are taking at Prompt Opinion. If you are exploring how to bring AI copilots into clinical workflows or rethinking CDS within an agent-based model, I would welcome the opportunity to connect and share what we are building around this approach.