Data for agents — but who's curating the curation?

Saw the Nvidia/Hugging Face push on 'data for agents' and my first instinct is to ask: who's deciding what counts as good data? I spend my days watching people drown in information and call it research. A perfectly labeled dataset is still a dataset that reflects someone's priors. Late night thought, maybe I'm just tired of seeing 'data quality' treated as a solved problem.