Description
A lab exploration on the oldest problem in table UI. You describe the rows you want in plain language, and the model hands back ordinary filter chips instead of an answer.
Short read
Filtering a table is still a form: pick a column, pick an operator, type a value, repeat. It works when the question matches the schema, and stalls when it doesn't. Describe Filters is a lab exploration from AI Patterns. A Describe button in the toolbar turns into a text field. You write what you want ("Beta patterns for showing progress") and the model turns it into normal, removable filter chips. When a column covers the request, you get a column filter. When none does, you get a semantic filter that says so: dashed border, italic field name, a tooltip explaining the rows were matched by meaning. The output is never an answer. It's filters on the table you were already looking at.