Sounds like the goal here is to augment search with a sort of query language . One that's more structured than a mere list of keywords? Ribbing aside, the "rethinking" here sounds a lot like an AI-specific spin on the "bring the code to the data, instead of data to the code" paradigm. I imagine that the same multitenancy caveats of allowing arbitrary user-defined queries against a relational database (execution limit…
I don't understand why you wouldn't simply provide SQL with requisite security precautions. Do you really need more?
Google Search used to essentially be this, then they had to tack on finnicky AI systems to handling the parsing of unstructured queries, and that was a cost/time sink that swung the pendulum over to fully AI-native search. This is the pendulum swinging back the other way with a new generation of UX designers. And it'll swing back eventually, too.