Agents in the Wild

This directory is no longer updated. Last update: November 16, 2026.

Cohere Health's policy agent cut the time to digitize a clinical policy from 2 h 15 min to 1 h 35 min

Cohere Health uses an agent to turn prior authorization policies from PDFs into structured, versioned data for health plans. Policy analysts review the output in a governed human-in-the-loop process.

Organization
Cohere Health
Industry
Healthcare software
When
Aug 2026
Evidence
Vendor case study

Reported outcome

Overall time spent on policy digitization fell 30 percent, from 2 hours 15 minutes to 1 hour 35 minutes per policy. Agent deployments went from 3–4 months to 2–6 weeks.

  • 2 hours 15 minutes to 1 hour 35 minutes (30 percent less)
    Time per policy
  • 3–4 months to 2–6 weeks
    Full agent deployment time

Caveats

Published by the company that sells the agent. Treat numbers as marketing until confirmed elsewhere.

  • Published on the AWS blog with Cohere Health co-authors. AWS sells the platform. Nobody outside checked the numbers.
  • A policy still takes 1 hour 35 minutes with people in the loop. The agent assists; it does not finish the work alone.
  • The post gives no accuracy or error rate for the digitized policies.
  • This agent structures policy documents. The post does not say it approves or denies any patient's request.
  • The knowledge graph and the 80 percent real-time approval goal are plans and targets, not results.

What the agent does

Health plans keep their prior authorization rules in documents and PDFs. Cohere Health's Policy Studio turns those rules into structured, machine-readable data, with the exact policy text beside a coded evidence layer and automatic version tracking.

The product runs one agent with several skills on Amazon Bedrock AgentCore. The agent reaches policy APIs as MCP tools through a gateway. Session memory keeps policy analysts' feedback, so teams can refine the output. The post says medical professionals stay responsible for clinical review where clinical judgment is needed.

Why this works

  • Clinical policy experts write and refine the skills. Domain knowledge is separate from the infrastructure.
  • A new skill must pass an evaluation suite, and data science approves it before production.
  • Each health plan's sessions run in isolation.
  • Versioned output makes it possible to see which policy version governed a decision.

What to copy

  • Put domain knowledge in skills that experts can edit without touching the runtime.
  • Gate every new skill with an evaluation suite and a named approver.
  • Keep the source text next to the structured output, with versions.

Starter prompt

Written by us, not by Cohere Health or AWS. A qualified reviewer must check every digitized rule before use.

Policy document: {policy_text}

Extract each coverage criterion as a row: criterion, exact source sentence, page, and logic (ALL of / ANY of).
Do not paraphrase the source sentence.
List anything unclear or contradictory for the policy analyst.
Do not decide any patient case.

Sources

  1. How Cohere Health digitizes clinical policies using Amazon Bedrock AgentCore — AWS, Aug 7, 2026

Added Nov 16, 2026 · Sources last checked Nov 16, 2026.

Get new use cases weekly

One short email. Sources included.

The newsletter is paused. Past issues stay in the archive.

Related use cases