From AI experiment to production: how Okta governs context for agents at scale
Okta rebuilt how it governs context for AI agents, using dbt as the single source of truth.
Overview
Many AI initiatives stall because "the data underneath isn't ready." Covers Okta's governed semantic layer (dbt + Snowflake) giving agents consistent, discoverable context from day one: scaling from PoC to production, portability/interoperability, and data-readiness standards for AI.
What you'll take away
- 01
Building the agent was never the hard part; what it could see and trust was — Okta found context fragmentation, no auditability, and undifferentiated instructions behind every failed AI answer.
- 02
A three-layer agent context framework, built entirely in dbt and shipped through normal pull requests, syncs automatically into Agents Schema so any agent with SQL access reads the same source of truth.
- 03
Fixing a wrong answer no longer means editing an opaque agent config: an analytics engineer merges one change to the context layer, and every agent that reads from it gets smarter immediately.
Session speaker
Pooja Crahen
Okta