Scaling Agents via Continual Pre-training - Explained Simply | ArXiv Explained