From AI experiment to production: how Okta governs context for agents at scale

Most AI pilots stall before they ship. The model works fine. The data underneath isn’t ready. At Okta, scaling from a proof-of-concept to something production-grade exposed a structural gap: agents built in silos, context fragmented, no shared foundation for meaning.

The missing piece wasn’t a better model. It was a governed, discoverable semantic layer that any agent could reason over from day one, built on dbt as the source of truth for definitions and Snowflake as the execution and discovery engine on top. Together, they turn a fragile demo into a system you’d trust in production. 

We’ll walk through the architecture, our decisions around portability and interoperability, and where we drew the line between production-ready and experimental. You’ll leave with a reusable pattern for making your data AI-ready at scale, and an honest look at what works today, what doesn’t, and where we’re going next.