Post by Sifflet

"Metadata" has been stretched into near-meaninglessness by a decade of data catalog marketing. Thomas Krakty brought some precision back at Signals25. For AI agents to reason reliably in an enterprise environment, they need three distinct layers: Technical metadata — where things live, how they connect, what the lineage looks like. Business metadata — what things mean, how metrics are defined, what the business glossary actually says. Trust metadata — data quality scores, PII classification, governance policies. The layer that tells an agent how much to rely on a given asset and what it's allowed to do with it. Miss any one of these and you're asking an agent to navigate an environment it can't read. Static documentation doesn't cut it. Metadata needs to be versioned, queryable, and observable — or it's not really metadata management at all. 🎬 Clip from our Signals25 rewind — full breakdown on the blog.

Video Content