we trained some new system one models (clef, clairic). they fit perfectly within our decade-old primitives like workers, our edge latency hypothesis for workers ai, and our ambition to be the agent cloud.