Discussion about this post

User's avatar
Foundation's Edge's avatar

The market history is the part most coverage skips. Standalone semantic layers died in 2022 because few companies felt real pain from inconsistent metrics, and the same vendors that let headless BI wither now ship semantic modeling natively. That reversal reads differently once the ablation behind it is public: Anthropic measured 21% accuracy for a frontier model pointed straight at the warehouse, and the same evals cleared 95% only behind a governed semantic boundary. The Supergrain history already names the open variable, since nobody publishes what curating those definitions costs, and that missing cost ledger decides the next round of this market. The same convergence and its three open variables, walked end to end: https://seldondance.substack.com/p/the-architecture-is-settled-the-economics

Joe Hovde's avatar

Love hearing about your experiences working with these! I always hated writing lookML haha

Ready for more?