also known as the semantic layer, previously known as the random queries in my BI tools
You're right that we haven't shared a lot of info on this of late! More soon, don't want to steal thunder. But I'm excited to get deep into the weeds with the entire community :D
I do, again, completely agree with you that the limitations you've identified are, well...limitations. That's ok. We're big fans of starting small and getting the community flywheel turning. v0 won't be suitable for every single team and every single use case, but as we have with dbt core over the course of 6 years, we'll continue to evolve and refine in partnership with the community.
Good round up! I think a semantic / metrics layer is still undervalued in the stack today. There's a ton of value in being able to aggregate or group data at query time rather than in a pipeline.
It appears to be difficult, if not impossible, to decouple a metrics layer from BI. Looking historically - Business Objects Universe had a front end, SSAS used Power BI or Excel, and Looker truly was built first as a semantic layer but continued into BI as a way to make it valuable - just to name a few examples. Today, Malloy has a built-in visualization library, Omni is building a hybrid of BI + metrics layer, and as you noted, Transform is heading in the BI direction.
History tells us that a headless metrics layer is both hard to build and hard to get adoption of (with maybe the exception of customer-facing data products). We have to consider what the value is of having your choice of front-end compared to just buying a tool that has BI/visualization built-in.
I'm really glad you used one of my favorite use cases as the example -- Defining activation for B2B SaaS products that allow one user to be part of mutliple workspaces!