The Failed Promises of Data Lineage: Why More Metadata Isn’t the Answer

​The data world doesn’t need more lineage graphs. Most are overly complex and lack meaningful context. After years of chasing “total visibility,” we’ve learned the hard way: more metadata doesn’t create more clarity.

​Collecting metadata isn’t the problem, understanding it is.

​The future of observability isn’t another graph, it’s a reasoning layer that connects lineage and telemetry into one continuous context. By unifying OpenLineage, OpenTelemetry and LLM reasoning, oleander builds a continuous feedback loop between data movement and system behavior giving data teams the context to detect, diagnose and adapt in real time.

​This talk explores how lineage has evolved, where it fell short and how an observability-first Telemetry Lake redefines it. We’ll discuss how oleander correlates telemetry, lineage and model context to reason about failures, propose fixes and continuously learn building what we call the always-on-call data engineer that never sleeps.

​Data lineage doesn’t need to map every dependency,  it needs a deeper understanding of the nodes and edges that drive impact, cost and key business decisions.

​Speaker

​Willy Lulciuc is the Co-Founder and CEO of Oleander, where he’s building the always on-call data engineer that never sleeps an agentic AI observability platform that transforms telemetry and lineage into real-time understanding. Before Oleander, Willy was a founding engineer at Datakin (acquired by Astronomer), where he co-created OpenLineage, now adopted by Google, Microsoft and Amazon. He also helped create Marquez, the open-source reference implementation for OpenLineage and serves as a TSC member for the LF AI & Data Foundation. At Oleander, he’s redefining data reliability for the AI era through open standards and intelligent observability.