Today's business intelligence tools:

• ■ Have clunky dev workflows. It's painful coordinating changes through dbt, semantic models, and dashboards. Simple changes can induce hours of manual fixes downstream.
• ■ Aren't keeping up in AI. Cursor and Claude Code are lightyears ahead of the best data agents. Software engineers are enjoying massive productivity gains. What about data analysts?
• ■ Are messy. Semantic models grow uncontrollably into a mess of duplicate logic and outdated one-offs. Usability declines for both humans and agents. Garbage in, garbage out.
• ■ Are hard to use. "Self serve" interfaces are still intimidating to average people. Configuring dashboards and visualizations takes an extraordinary amount of clicks. More friction = less insights.

We believe that the AI era presents an opportunity to completely rethink how software is built. A modern BI solution should have the following traits:

• ■ Lives in your repo. Your semantic model and dashboards should be backed by code in the same repo where you already do other data work. This means that changes can percolate along the stack in a single commit. Dependencies are easily traceable, and dev stages move in lockstep.
• ■ Compatible with state-of-the-art agents. Data professionals should be able to leverage the best coding agents for automating data work. All data actions should be available as CLIs so that agents can learn, reason, and iterate.
• ■ A semantic layer that stays healthy over time. The semantic layer should have zero redundancy with your dbt code and should be constantly maintained and documented by background agents.
• ■ Agentic dashboard creation for self-service. Building dashboards should be as easy as simply speaking. And the data logic should be governed by the semantic model so that the AI stays on the rails.