The catch Not all semantic layers can do this.
The reasoning, the lineage, the AQL above — every trust mechanism relies on the layer underneath going far enough. Most semantic layers stop short.
Above the ceiling — the questions your business actually asks
Cohort retention by signup month Period-over-period across regions Ratios across grains Nested aggregation Running totals by segment
↳ handled off-platform — spreadsheets, workbook calcs, ad-hoc SQL. Ungoverned. Unverifiable.
Where conventional semantic layers stop —
and where governance stops with them.
The same ceiling your AI agent inherits the moment you ship it.
Below the ceiling — the easy half
Revenue last month Top 10 channels Slice & filter by region Sum, count, average
↳ any semantic layer handles these — and governs them.
Above the ceiling, your AI hits a wall — and the work leaks off-platform, ungoverned. Below it, anyone's tool works. The difference is how high your ceiling goes.
That's why we built AML and AQL — to push the ceiling, and keep governance with it.