How to use Unknown to submit prior authorizations
Silna built insurance-portal agents with isolated browser sessions, persisted clinical-data pipelines, recorded actions and human handoffs for uncertain situations.
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Design [prior authorization portal workflow] using isolated, provider-scoped browser sessions. Extract and validate [required clinical fields] from [source documents] before submission. Restrict navigation to [portal domain], preserve recordings and artifacts, and hand off to [human reviewer] when confidence is insufficient.
Reach people exploring AI agents
See placements →What happened
Silna reports that browser agents submit nearly twenty percent of its prior authorization volume and that this volume scaled sevenfold over the past year.
Reported by bgschulman31 on BlogHow the agent did it
- Evaluated vendors and selected self-hosted Kasm Workspaces.
- Built isolated provider-scoped browser session management.
- Connected Playwright to running browser sessions through CDP.
Where a person stepped in
Silna supports agent-to-human handoffs when confidence is insufficient and observes changes through real-time supervised production experiments.
A person approved each step before the agent continued.
Silna Health describes browser agents that carry out prior authorization workflows on insurance portals. The agents log in with two-factor authentication, navigate legacy interfaces, enter clinical data and record approvals in the company's system. The source says they submit nearly twenty percent of its prior authorization volume and that this volume scaled sevenfold over the preceding year.
The team evaluated remote-browser and virtual-machine vendors before choosing self-hosted Kasm Workspaces. They built a session lifecycle that provisions containers, injects policies, syncs files and archives recordings after work completes. Sessions are provider-scoped through database row-level security, and Chrome policies restrict browsing to the target portal domain. Playwright connects to running sessions through Chrome DevTools Protocol.
Before browser submission, clinical data must be extracted from PDFs, validated and connected to its source documents. Silna built a Postgres-backed orchestration layer over Dramatiq, called Pipelines, to persist steps, results and failures. Independent clinical-field extractions can run in parallel, while submission waits for all required work.
The infrastructure supports human handoffs when agents lack confidence, and changes are observed through supervised production experiments. Who it's for: healthcare operations teams developing auditable portal automation with explicit preparation and human escalation paths.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Blog · bgschulman31
Frequently asked questions
What happens before an agent opens the portal form?
The system extracts clinical data from PDFs, validates the extractions, identifies their locations in source documents and prepares the information needed for submission.
How are browser sessions made auditable?
Sessions have full video recordings and intercepted file transfers. At completion, artifacts sync to S3 with metadata, and recordings are archived alongside the associated task.
More details
Remaining steps
- Built persisted pipelines for clinical extraction and validation.
- Ran portal submissions with human handoffs and supervised experiments.
About this account
Agents handle portal login, navigation, clinical-data entry and approval recording after preparatory pipelines complete.
Platform: Blog. Attribution: https://engineering.silnahealth.com/posts/autonomous-browser-agents-enterprise/. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: bgschulman31 on Blog
Source published 24 Feb 2026
Source checked 14 Oct 2026
Tools and topics
Kasm Workspaces · Playwright · Chrome DevTools Protocol · Postgres · Dramatiq · S3
Prior authorization · Browser automation · Auditability
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.