Goodbye Brittle Dashboards, Hello Conversational BI

I spent a significant part of my career building OLAP cubes.

Carefully designed star schemas. Meticulously maintained conformed dimensions. Aggregation tables that shaved milliseconds off query times. That infrastructure was real work.

Today, it's increasingly obsolete.

The Problem with Prepared BI

Traditional BI worked like this: a data engineer pre-aggregated answers into a semantic model, and surfaced those models through tools like Tableau or Power BI.

This created a fundamental bottleneck: the rate at which analysts could ask new questions was throttled by the rate at which engineers could update the model. Miss a dimension? Analysts hit a wall. Wrong join in the semantic layer? A month of bad reports before anyone noticed.

And don't get me started on Tableau's access control. Brittle doesn't begin to cover it.

The Metadata-Driven Alternative

What's replacing this model is fundamentally different. Instead of pre-building answers, you build a rich metadata layer — meaning, relationships, quality, lineage — and let AI navigate it dynamically.

A knowledge graph, essentially. Not a static OLAP cube, but a dynamic graph of entities and relationships that an AI system can traverse in response to natural language questions.

"Show me customers who churned in Q3 but had high NPS scores in Q2."

In the old world, that question might require a new join in the semantic model and two weeks of development. In the new world, it's a conversation.

GCP Is Leading This Shift

Google Cloud Platform has been aggressive about integrating conversational AI directly into the data stack. BigQuery's natural language querying, Looker's AI features, Gemini integrated throughout — these aren't experiments. They're production features used by real organizations today.

The move from Tableau to GCP-native BI is also a move from brittle infrastructure to platform-integrated tooling. Access control is handled at the data layer (IAM, column-level security in BigQuery), not bolted on in a BI tool with its own user management nightmare.

What This Means for Data Engineers

Less time on: building semantic models, managing BI tool infrastructure, writing calculated fields.

More time on: building rich metadata catalogs, defining entity relationships and data lineage, ensuring data quality so AI queries return trustworthy results, managing security at the data layer.

The knowledge graph is only as good as the metadata behind it. That's where your expertise becomes irreplaceable.