Builder Guide
Harness Packs
Test AI behavior configurations against scored scenarios before they touch real customers.
Audience
Builders using AI features
Purpose
Pre-release AI quality gate
Related
AI assistant
Before an AI behavior configuration serves your customers, you can run it through a Harness Pack: a set of scenarios with expected outcomes, scored automatically. Configurations that don't clear the bar don't ship.
Why this exists
AI behavior is configuration, and configuration drifts. Harness Packs make quality a gate instead of a hope — the same way you wouldn't deploy code without tests.
Working with packs
- Choose or author a pack of scenarios relevant to your use case.
- Run your candidate configuration against it; each scenario is scored.
- Review failures, adjust, re-run. Promote only when the pack passes.
Capability review: 2026-09-14. For exact current technical availability, use the generated API Map and first-class module inventory.
Did this page answer your question?Tell us what’s missing