Cohere Health Digitizes Clinical Policies 30% Faster with Amazon Bedrock AgentCore

Title

Cohere Health Digitizes Clinical Policies 30% Faster with Amazon Bedrock AgentCore

Content

Cohere Health, a clinical intelligence company that powers health plan operations, built Cohere Policy Studio using Amazon Bedrock AgentCore, reducing policy digitization time by 30 percent — from 2 hours 15 minutes to 1 hour 35 minutes per policy — and cutting AI agent deployment from months to weeks, while supporting sub-second policy retrieval and real-time updates across connected decisioning products. The driver was prior authorization, the approval process health plans require before covering medical services or medications — one of the most manual processes in healthcare because the governing policies are trapped in static, unstructured formats (documents and PDFs) that resist automation. Policy content varies by clinical area, geography, line of business, and health plan, and evolves as medicine advances. Cohere Health identified three challenges: CMS regulations requiring API-based electronic prior authorization by January 2027; America's Health Insurance Plans (AHIP) commitments requiring 80 percent real-time approvals for electronic submissions; and technical architecture demands of ingesting multiple input formats while producing different policy representations for different downstream consumers, each with its own feedback loop. The team completed three tasks to build Policy Studio: deployed AgentCore Runtime with AgentCore Gateway and AgentCore Memory for a full agentic system using LangChain; configured AgentCore Gateway to fetch tools and skills; and wrote skills with clinical policy experts, evaluating them through Cohere Health's standardized observability process based on Arize AI. Three architectural patterns made it scalable: reusable Amazon ECR base images that separate the stable runtime environment from team-specific agent_config.yaml (with options for memory modes, storage strategies, prompt caching, and LiteLLM reverse proxy configuration); AgentCore Gateway consolidating tools — AWS Lambda functions and internal APIs exposed as Model Context Protocol (MCP) tools — behind a single authenticated endpoint; and a modular skills framework built on the Agent Skills open standard so clinical policy experts can author and refine skills directly without rebuilding the agent. AgentCore Runtime's secure microVM isolation enforces strict data isolation between health plan tenants with dedicated compute, memory, and filesystem resources per session. A knowledge graph designed into the system identifies relevant connections between digitized policies, flags potential conflicts, and suggests reusable patterns — learning from policy structures across clinical areas to accelerate time-to-deployment for new indication types from days to hours. Decisioning engines query the graph via natural language or FHIR resources, retrieving policy fragments with full provenance and version history, while maintaining bidirectional links between CMS requirements, AHIP commitments, and internal policy representations. The graph supports helping health plans achieve 80 percent of electronic prior authorization approvals in real time, with comprehensive indication coverage without proportional engineering effort, automated conflict detection, and sub-second policy retrieval. Cohere Health has digitized thousands of policies to date, with clinical reviewers remaining responsible for clinical judgment within a governed, human-in-the-loop process.