Building AgentOS: Why I’m Building the AWS Lambda for Insurance Claims
Published: (April 21, 2026 at 04:42 AM EDT)
2 min read
Source: Dev.to
Source: Dev.to
The Problem: The 15-Day Bottleneck
The average P&C (Property & Casualty) claim still takes 15+ days to process, involves 10+ manual touchpoints, and costs carriers a fortune in Loss Adjustment Expenses (LAE). Legacy systems are monoliths—integrating a new “AI tool” usually takes 18 months of enterprise red tape.
The Vision: AgentOS
I’m building AgentOS. It’s not an “AI wrapper.” It’s a modular, AI‑native orchestration layer—think of it as AWS Lambda for claims. Instead of one giant LLM trying to do everything, AgentOS deploys a fleet of specialized agents:
- Intake Agent – Converts messy FNOL (First Notice of Loss) data into structured JSON.
- Policy Agent – Cross‑references claims against complex coverage documents in milliseconds.
- Fraud Shield – An agentic pattern‑matcher that flags anomalies before they hit the payout stage.
The Goal: 70% Straight‑Through Processing (STP)
Targeting low‑complexity claims such as auto‑glass, the aim is to achieve 70 % STP for these cases.
Technical Critiques Requested
- Pricing model: Is a usage‑based “Margin‑on‑Claim” model better than a standard SaaS seat license for infrastructure?
- State persistence: How would you handle state persistence for agents that need to wait 24 hours for a human‑approval gate?