- https://cube.dev/ # THE Agentic Analytics Platform
The “Human” Semantic Layer: Jayatillake argues that every company has a semantic layer; if it isn’t codified in software, it exists inefficiently within the heads of data analysts. If an analyst constantly has to write custom SQL to answer stakeholder questions because they are the only ones who know the data model, they are acting as a “human semantic layer”.
Decoupling from BI Tools: A critical insight is the danger of defining semantics within specific Business Intelligence (BI) tools (like Looker or PowerBI).
The Problem: Logic defined in a BI tool’s proprietary language (DSL) is locked into that platform. If you want to use that same logic for an ML pipeline, a React app, or a different BI tool, you must duplicate the code, leading to “different versions of the truth”.
The Solution: The semantic layer should be abstracted out and hosted separately (e.g., using open-source tools like Cube). This “headless” approach allows a single definition of metrics to serve BI tools, AI agents, ML pipelines, and custom applications simultaneously