What is built today, and what comes next.

This page lists what the Gen22 platform does today and what is on the roadmap. If a claim anywhere else on this site goes beyond this page, this page is the one that holds. Ask us to show you any item on a call.

Built today

Architecture checks

Automated checks run on every push and in CI. They refuse any change that would:

• call an AI model without recording the decision;
• run a command or query without declaring its authorization;
• bypass the system that records every request;
• read or write data outside approved entry points;
• skip tenant data isolation.

The checks apply whether a person or an AI agent wrote the code. They enforce the platform's architecture rules.

The record

• One connected record. Every database row is tied to the request that caused it, the logs and traces around it and, where there is one, the AI decision that produced it and the human approval that confirmed it.
• AI decisions. What the model saw, which model and version answered, what it recommended and what it cost.
• Human review. Workflows pause for a licensed reviewer. The model's recommendation and the reviewer's decision are stored as separate fields, so an override is a recorded fact, with notes.
• Third-party vendors. Every outside call records the vendor's engine, version and ruleset. Failures and retries stay on the record.
• Permission checks. Every check is logged, including refusals.
• Health data. Fields marked as PHI are masked automatically in every log and trace.

Audit packet

• One view answers seven questions: who asked, what ran, what the AI saw, which third party answered, what it cost, who approved, what changed.
• Each answer is traced to the database table and row it came from. A missing answer is shown as missing.
• Exportable as JSON or PDF.
• Audit rows are append-only: no operating identity can delete selected rows. Retention is configured per record type.

Operations

• Live system map, generated from the code, giving a current inventory of what the system does.
• Alerting. Thresholds are declared in code, and fired alerts are routed automatically by severity.
• One model interface. Anthropic is the provider wired today. Adding another provider is an adapter, not a rebuild.
• Deployment. Designed to run in the carrier's own environment. Google Cloud is the environment set up today.

On the roadmap

None of these are shipped.

• Self-maintaining systems. AI agents pick up alerts, follow the record to the cause and propose a fix for a human to approve. Today the alert delivery is built; the agent step is designed, not built.
• Workflow templates for common insurance work, such as submission intake, rater replacement and bordereaux.
• Using the record before acting: the accumulated record as input to a decision, not only for audits after it.

Company

Gen22 is pre-revenue with a working product and no customers yet. We are recruiting founding design partners. The dental predetermination demo on the home page uses synthetic data only.