Cohere Health digitizes clinical policies using AWS Bedrock agents

Prior authorization is the approval process health plans require before covering certain medical services or medications. It remains one of healthcare's most manual processes because the policies that govern it are trapped in static, unstructured formats—PDFs, documents, and manual records—that resist automation. This content is at the core of day-to-day clinical operations affecting hundreds of millions of patients each year, yet the policy content varies by clinical area, geography, line of business, and health plan, and evolves as medicine and technology advance. Historically, health plans did not have a systematic way to manage, analyze, and optimize them.

Cohere Health, a clinical intelligence company that powers health plan operations, built Cohere Policy Studio using Amazon Bedrock AgentCore to address this challenge. The application uses a flexible, multi-tenant agentic architecture to accelerate policy digitization with extensive workflow management and automatic version tracking. Three regulatory and technical drivers shaped the solution: Centers for Medicare & Medicaid Services (CMS) regulations require health plans to support API-based electronic prior authorization by January 2027; America's Health Insurance Plans (AHIP) commitments require health plans to achieve 80 percent real-time approvals for electronic prior authorization submissions, with each line of business having unique requirements; and the technical architecture needed to ingest multiple input formats and produce different representations of each policy for different downstream consumers, each with its own feedback loop.

Cohere Health's architecture separates the stable runtime environment from team-specific configurations using a two-tier deployment approach. The FROM line pulls a shared base image containing the LangChain agent framework and common dependencies; the COPY line adds team-specific agent configuration files that control memory modes (stateless or persistent conversation history), storage strategies (full trace for correction workflows or conversation-only for clean history), session context caching, prompt caching, flexible tool configuration, model configuration with Amazon Bedrock, and LiteLLM configuration as a reverse proxy. AgentCore Gateway consolidates multiple tool types—AWS Lambda functions for fetching skills and documents, and internal APIs maintained across different teams—behind a single authenticated endpoint, allowing teams to add new tools without redeploying the agent. The gateway invokes a Lambda function for each tool request, routing to the correct handler based on the tool name, and uses environment variables for configuration such as the S3 bucket and prefix for skill definitions.

Skills development follows a structured workflow where clinical policy experts author and refine new skills directly, ensuring the system supports policy workflows grounded in expert review and governance. Each skill is evaluated for accuracy, completeness, and consistency against reference datasets containing ground truth outputs. When a skill fails, the team analyzes the failure mode and iterates on the skill definition before retesting. After passing the evaluation suite, data science reviews results against acceptance criteria and approves the skill for production deployment. In production, Arize AI tracks effectiveness metrics, and clinical policy analysts annotate sample outputs to catch errors that automated metrics miss. The team monitors for skill degradation over time and uses these data points to prioritize optimization work. By decoupling domain expertise from infrastructure through modular, versioned skill definitions, teams deploy new capabilities without rebuilding the agent, creating a pathway for rapid deployment velocity across multiple health plans while preserving transparency, version control, and human oversight.