Most data platforms document their guarantees and enforce none of them. A table is described as daily, a lineage diagram sits in a wiki, and quality is a dashboard someone checks after an incident. Models and agents built on that foundation inherit every undocumented change upstream, and the failure surfaces as a wrong answer rather than a broken pipeline.

A data contract makes those guarantees testable. This keynote covers what belongs in one: freshness windows, schema and semantic stability, lineage that resolves to a source, and quality checks that run inside the pipeline rather than alongside it. It also covers what happens at the boundary when a producer breaks a contract a consumer depends on.

The session closes on rolling contracts out incrementally across a large, regulated enterprise: which datasets to cover first, how to enforce without stopping delivery, and how contract coverage changes what an AI system can be trusted to do on top of it.