A powerful moment in our webinar demo was when Artyom asked for "prior month's revenue" and "month-over-month growth"—metrics that were not pre-defined in the data model. For anyone who has worked with traditional BI or semantic layers, this is a familiar pain point. If a metric doesn't exist in the model, the workflow grinds to a halt. You have to file a ticket with the data team, wait for them to add the logic, and then wait for a deployment. This friction between the speed of business questions and the pace of data modeling is a major source of frustration.

Cube D3 is designed to solve this by making ad-hoc analysis an interactive, real-time experience, without sacrificing the governance that the semantic layer provides.

How It Works: Composing with Governed Building Blocks

The magic behind this capability lies in the fact that D3’s agents generate Semantic SQL. Because Semantic SQL is a superset of standard SQL, the agent can use common SQL functions and expressions to build new calculations on top of your existing, governed metrics. It’s not writing queries against raw, untrusted tables; it’s composing new logic using trusted, pre-defined building blocks.

Let's break down the "month-over-month growth" example:

  1. The Core Metric: The agent starts with the orders.revenue measure, which is a governed object from the semantic layer. It knows this metric is reliable and its definition is consistent.
  2. The Ad-Hoc Calculation: To get the prior month's revenue, the agent applies a standard LAG() window function to the orders.revenue measure within its Semantic SQL query. It’s performing a well-understood SQL operation on a trusted input.
  3. The Final Metric: To calculate the final month-over-month growth percentage, the agent then uses the results of the first two steps to perform the division.

The beauty of this approach is that the entire ad-hoc analysis is still anchored to the single source of truth. The agent isn’t inventing raw data or making up logic; it’s simply performing new math on trusted data, all within the secure execution environment of the semantic layer.

Image

From Ad-Hoc to Governed: The AI Data Engineer Copilot

This creates what we can think of as a multi-layered semantic model. You have the core, durable model managed by the data team, and now you have a flexible, "just-in-time" modeling layer where analysts can explore.

But the lifecycle doesn't end there. As we saw with the "Average Order Value" example, D3 provides a path to promote valuable ad-hoc logic. An analyst can work with the AI Data Engineer agent, which acts as a copilot for your semantic layer. You can simply ask it to "save my average order value calculation to the orders cube." The agent will generate the correct YAML or JavaScript, add descriptions, and present it for review.

This closes the loop. An exploration that started as a "quick win" can become an "enterprise standard" without a lengthy development cycle. It’s how you go from ad-hoc insight to governed, reusable logic, ensuring your semantic model evolves with the needs of the business.

[Screenshot with UI Mock-up - Promoting a Metric]

Next up: We'll discuss how D3 builds institutional knowledge with features like Certified Queries and Memories.

Interested in learning more about D3? Join the waitlist for Cube D3 today.