Reliance Global Group (NASDAQ: EZRA) on Monday launched a proprietary AI agent built for browser automation in regulated insurance back offices, choosing to build its own infrastructure rather than deploy a general-purpose agent framework like OpenClaw or Claude Code.
The agent, announced via GlobeNewswire, automates repetitive portal tasks that insurance teams perform daily: service requests, endorsements, quote pulls, status checks, and document retrieval. The design philosophy inverts the autonomy-first approach of most agent frameworks. Where general-purpose agents maximize what they can do independently, Reliance’s agent maximizes what it can do while keeping the organization in control.
How the Controls Work
The architecture enforces five constraints that distinguish it from general-purpose agents:
Policy-enforced action blocking. Every browser action passes through a centralized enforcement layer. Irreversible actions like submit, issue, or bind are blocked entirely and reserved for human employees.
Save-and-park review. The agent never finalizes a transaction. Completed work is saved within the carrier portal and held for employee approval. Each run is independently scored against its own screenshots and event logs, not the agent’s self-report, and flagged for review if anything falls short.
Secure credential handling. Portal credentials stay under agency control, retrieved securely at runtime and never exposed to the AI model or recorded in logs.
Risk classification in code. Every task type carries a risk classification assigned programmatically, not by the AI. Verification checks catch potential errors including unintended pricing changes.
Audit-grade logging. Every workflow produces step-by-step event logs with screenshots, designed for reporting, review, and compliance oversight under GLBA, NAIC Model 668, and NY DFS 500.
“Most AI agents built for the browser are designed to maximize what they can do independently. We built ours to maximize what it can do while keeping organizations in control,” said Zack Wilder, Reliance’s CTO, who previously led engineering at Coinbase and Capital One.
Commercialization Plans
Reliance plans to deploy the agent across its growing network of acquired insurance agencies first, then commercialize externally. The pricing structure includes a free developer tier, Team workspace, and Enterprise plan with self-hosted and virtual private cloud (VPC) options. Each tier includes the core security controls, human review, and audit reporting framework.
COO Judah Korman framed the agent as an acquisition integration tool: “Every agency we bring onto the platform comes with thousands of hours of repetitive portal work. This agent allows us to absorb that work without scaling headcount while maintaining the controls required in insurance operations.”
Why Custom Over General-Purpose
The decision to build proprietary infrastructure rather than adopt an existing agent framework reflects a tension in regulated industries. General-purpose agent platforms optimize for flexibility and developer experience. Regulated environments need the opposite: constrained action spaces, deterministic enforcement, and audit trails that satisfy specific regulatory frameworks.
Reliance’s approach trades the breadth of a platform like OpenClaw for domain-specific compliance guarantees baked into the agent’s architecture. The company expects the technology to expand beyond insurance back offices into broader regulated enterprise use cases as it adds workflows and carrier portal integrations.